1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
48 #include "mac-learning.h"
52 #include "ofp-print.h"
54 #include "ofproto/netflow.h"
55 #include "ofproto/ofproto.h"
56 #include "ovsdb-data.h"
58 #include "poll-loop.h"
62 #include "socket-util.h"
63 #include "stream-ssl.h"
65 #include "system-stats.h"
70 #include "vswitchd/vswitch-idl.h"
71 #include "xenserver.h"
73 #include "sflow_api.h"
75 VLOG_DEFINE_THIS_MODULE(bridge);
77 COVERAGE_DEFINE(bridge_flush);
78 COVERAGE_DEFINE(bridge_process_flow);
79 COVERAGE_DEFINE(bridge_process_cfm);
80 COVERAGE_DEFINE(bridge_process_lacp);
81 COVERAGE_DEFINE(bridge_reconfigure);
82 COVERAGE_DEFINE(bridge_lacp_update);
90 struct dst builtin[32];
95 static void dst_set_init(struct dst_set *);
96 static void dst_set_add(struct dst_set *, const struct dst *);
97 static void dst_set_free(struct dst_set *);
100 /* These members are always valid. */
101 struct list port_elem; /* Element in struct port's "ifaces" list. */
102 struct port *port; /* Containing port. */
103 char *name; /* Host network device name. */
104 tag_type tag; /* Tag associated with this interface. */
105 long long delay_expires; /* Time after which 'enabled' may change. */
107 /* These members are valid only after bridge_reconfigure() causes them to
109 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
110 int dp_ifidx; /* Index within kernel datapath. */
111 struct netdev *netdev; /* Network device. */
112 bool enabled; /* May be chosen for flows? */
113 bool up; /* Is the interface up? */
114 const char *type; /* Usually same as cfg->type. */
115 const struct ovsrec_interface *cfg;
117 /* LACP information. */
118 uint16_t lacp_priority; /* LACP port priority. */
121 #define BOND_MASK 0xff
123 struct iface *iface; /* Assigned iface, or NULL if none. */
124 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
125 tag_type tag; /* Tag for bond_entry<->iface association. */
129 BM_TCP, /* Transport Layer Load Balance. */
130 BM_SLB, /* Source Load Balance. */
131 BM_AB /* Active Backup. */
134 #define MAX_MIRRORS 32
135 typedef uint32_t mirror_mask_t;
136 #define MIRROR_MASK_C(X) UINT32_C(X)
137 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
139 struct bridge *bridge;
142 struct uuid uuid; /* UUID of this "mirror" record in database. */
144 /* Selection criteria. */
145 struct shash src_ports; /* Name is port name; data is always NULL. */
146 struct shash dst_ports; /* Name is port name; data is always NULL. */
151 struct port *out_port;
155 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
157 struct bridge *bridge;
158 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
161 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
162 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
163 * NULL if all VLANs are trunked. */
164 const struct ovsrec_port *cfg;
167 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
168 long long int miimon_interval; /* Miimon status refresh interval. */
169 long long int miimon_next_update; /* Time of next miimon update. */
171 /* An ordinary bridge port has 1 interface.
172 * A bridge port for bonding has at least 2 interfaces. */
173 struct list ifaces; /* List of "struct iface"s. */
174 size_t n_ifaces; /* list_size(ifaces). */
177 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
178 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
179 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
180 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
181 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
182 long long int bond_next_fake_iface_update; /* Time of next update. */
184 /* LACP information. */
185 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
186 bool lacp_active; /* True if LACP is active */
187 bool lacp_fast; /* True if LACP is in fast mode. */
188 uint16_t lacp_priority; /* LACP system priority. */
190 /* SLB specific bonding info. */
191 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
192 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
193 long long int bond_next_rebalance; /* Next rebalancing time. */
195 /* Port mirroring info. */
196 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
197 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
198 bool is_mirror_output_port; /* Does port mirroring send frames here? */
202 struct list node; /* Node in global list of bridges. */
203 char *name; /* User-specified arbitrary name. */
204 struct mac_learning *ml; /* MAC learning table. */
205 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
206 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
207 const struct ovsrec_bridge *cfg;
209 /* OpenFlow switch processing. */
210 struct ofproto *ofproto; /* OpenFlow switch. */
212 /* Kernel datapath information. */
213 struct dpif *dpif; /* Datapath. */
214 struct hmap ifaces; /* Contains "struct iface"s. */
217 struct hmap ports; /* "struct port"s indexed by name. */
218 struct shash iface_by_name; /* "struct iface"s indexed by name. */
221 bool has_bonded_ports;
226 /* Port mirroring. */
227 struct mirror *mirrors[MAX_MIRRORS];
230 /* List of all bridges. */
231 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
233 /* OVSDB IDL used to obtain configuration. */
234 static struct ovsdb_idl *idl;
236 /* Each time this timer expires, the bridge fetches systems and interface
237 * statistics and pushes them into the database. */
238 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
239 static long long int stats_timer = LLONG_MIN;
241 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
242 static void bridge_destroy(struct bridge *);
243 static struct bridge *bridge_lookup(const char *name);
244 static unixctl_cb_func bridge_unixctl_dump_flows;
245 static unixctl_cb_func bridge_unixctl_reconnect;
246 static int bridge_run_one(struct bridge *);
247 static size_t bridge_get_controllers(const struct bridge *br,
248 struct ovsrec_controller ***controllersp);
249 static void bridge_reconfigure_one(struct bridge *);
250 static void bridge_reconfigure_remotes(struct bridge *,
251 const struct sockaddr_in *managers,
253 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
254 static void bridge_fetch_dp_ifaces(struct bridge *);
255 static void bridge_flush(struct bridge *);
256 static void bridge_pick_local_hw_addr(struct bridge *,
257 uint8_t ea[ETH_ADDR_LEN],
258 struct iface **hw_addr_iface);
259 static uint64_t bridge_pick_datapath_id(struct bridge *,
260 const uint8_t bridge_ea[ETH_ADDR_LEN],
261 struct iface *hw_addr_iface);
262 static uint64_t dpid_from_hash(const void *, size_t nbytes);
264 static unixctl_cb_func bridge_unixctl_fdb_show;
265 static unixctl_cb_func qos_unixctl_show;
267 static void bond_init(void);
268 static void bond_run(struct port *);
269 static void bond_wait(struct port *);
270 static void bond_rebalance_port(struct port *);
271 static void bond_send_learning_packets(struct port *);
272 static void bond_enable_slave(struct iface *iface, bool enable);
274 static void port_run(struct port *);
275 static void port_wait(struct port *);
276 static struct port *port_create(struct bridge *, const char *name);
277 static void port_reconfigure(struct port *, const struct ovsrec_port *);
278 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
279 static void port_destroy(struct port *);
280 static struct port *port_lookup(const struct bridge *, const char *name);
281 static struct iface *port_lookup_iface(const struct port *, const char *name);
282 static struct iface *port_get_an_iface(const struct port *);
283 static struct port *port_from_dp_ifidx(const struct bridge *,
285 static void port_update_bonding(struct port *);
286 static void port_update_lacp(struct port *);
288 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
289 static void mirror_destroy(struct mirror *);
290 static void mirror_reconfigure(struct bridge *);
291 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
292 static bool vlan_is_mirrored(const struct mirror *, int vlan);
294 static struct iface *iface_create(struct port *port,
295 const struct ovsrec_interface *if_cfg);
296 static void iface_destroy(struct iface *);
297 static struct iface *iface_lookup(const struct bridge *, const char *name);
298 static struct iface *iface_find(const char *name);
299 static struct iface *iface_from_dp_ifidx(const struct bridge *,
301 static void iface_set_mac(struct iface *);
302 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
303 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
304 static void iface_update_cfm(struct iface *);
305 static void iface_refresh_cfm_stats(struct iface *iface);
306 static void iface_update_carrier(struct iface *);
307 static bool iface_get_carrier(const struct iface *);
309 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
311 static void shash_to_ovs_idl_map(struct shash *,
312 char ***keys, char ***values, size_t *n);
314 /* Hooks into ofproto processing. */
315 static struct ofhooks bridge_ofhooks;
317 /* Public functions. */
319 /* Initializes the bridge module, configuring it to obtain its configuration
320 * from an OVSDB server accessed over 'remote', which should be a string in a
321 * form acceptable to ovsdb_idl_create(). */
323 bridge_init(const char *remote)
325 /* Create connection to database. */
326 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
328 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
329 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
330 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
332 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
334 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
335 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
337 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
338 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
339 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
341 /* Register unixctl commands. */
342 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
343 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
344 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
346 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
355 struct bridge *br, *next_br;
357 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
360 ovsdb_idl_destroy(idl);
363 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
364 * but for which the ovs-vswitchd configuration 'cfg' is required. */
366 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
368 static bool already_configured_once;
369 struct svec bridge_names;
370 struct svec dpif_names, dpif_types;
373 /* Only do this once per ovs-vswitchd run. */
374 if (already_configured_once) {
377 already_configured_once = true;
379 stats_timer = time_msec() + STATS_INTERVAL;
381 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
382 svec_init(&bridge_names);
383 for (i = 0; i < cfg->n_bridges; i++) {
384 svec_add(&bridge_names, cfg->bridges[i]->name);
386 svec_sort(&bridge_names);
388 /* Iterate over all system dpifs and delete any of them that do not appear
390 svec_init(&dpif_names);
391 svec_init(&dpif_types);
392 dp_enumerate_types(&dpif_types);
393 for (i = 0; i < dpif_types.n; i++) {
396 dp_enumerate_names(dpif_types.names[i], &dpif_names);
398 /* Delete each dpif whose name is not in 'bridge_names'. */
399 for (j = 0; j < dpif_names.n; j++) {
400 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
404 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
413 svec_destroy(&bridge_names);
414 svec_destroy(&dpif_names);
415 svec_destroy(&dpif_types);
418 /* Callback for iterate_and_prune_ifaces(). */
420 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
422 if (!iface->netdev) {
423 /* We already reported a related error, don't bother duplicating it. */
427 if (iface->dp_ifidx < 0) {
428 VLOG_ERR("%s interface not in %s, dropping",
429 iface->name, dpif_name(br->dpif));
433 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
434 iface->name, iface->dp_ifidx);
438 /* Callback for iterate_and_prune_ifaces(). */
440 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
441 void *aux OVS_UNUSED)
443 /* Set policing attributes. */
444 netdev_set_policing(iface->netdev,
445 iface->cfg->ingress_policing_rate,
446 iface->cfg->ingress_policing_burst);
448 /* Set MAC address of internal interfaces other than the local
450 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
451 iface_set_mac(iface);
457 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
458 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
459 * deletes from 'br' any ports that no longer have any interfaces. */
461 iterate_and_prune_ifaces(struct bridge *br,
462 bool (*cb)(struct bridge *, struct iface *,
466 struct port *port, *next_port;
468 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
469 struct iface *iface, *next_iface;
471 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
472 if (!cb(br, iface, aux)) {
473 iface_set_ofport(iface->cfg, -1);
474 iface_destroy(iface);
478 if (!port->n_ifaces) {
479 VLOG_WARN("%s port has no interfaces, dropping", port->name);
485 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
486 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
487 * responsible for freeing '*managersp' (with free()).
489 * You may be asking yourself "why does ovs-vswitchd care?", because
490 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
491 * should not be and in fact is not directly involved in that. But
492 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
493 * it has to tell in-band control where the managers are to enable that.
494 * (Thus, only managers connected in-band are collected.)
497 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
498 struct sockaddr_in **managersp, size_t *n_managersp)
500 struct sockaddr_in *managers = NULL;
501 size_t n_managers = 0;
502 struct shash targets;
505 /* Collect all of the potential targets from the "targets" columns of the
506 * rows pointed to by "manager_options", excluding any that are
508 shash_init(&targets);
509 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
510 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
512 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
513 shash_find_and_delete(&targets, m->target);
515 shash_add_once(&targets, m->target, NULL);
519 /* Now extract the targets' IP addresses. */
520 if (!shash_is_empty(&targets)) {
521 struct shash_node *node;
523 managers = xmalloc(shash_count(&targets) * sizeof *managers);
524 SHASH_FOR_EACH (node, &targets) {
525 const char *target = node->name;
526 struct sockaddr_in *sin = &managers[n_managers];
528 if ((!strncmp(target, "tcp:", 4)
529 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
530 (!strncmp(target, "ssl:", 4)
531 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
536 shash_destroy(&targets);
538 *managersp = managers;
539 *n_managersp = n_managers;
543 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
545 struct shash old_br, new_br;
546 struct shash_node *node;
547 struct bridge *br, *next;
548 struct sockaddr_in *managers;
551 int sflow_bridge_number;
553 COVERAGE_INC(bridge_reconfigure);
555 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
557 /* Collect old and new bridges. */
560 LIST_FOR_EACH (br, node, &all_bridges) {
561 shash_add(&old_br, br->name, br);
563 for (i = 0; i < ovs_cfg->n_bridges; i++) {
564 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
565 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
566 VLOG_WARN("more than one bridge named %s", br_cfg->name);
570 /* Get rid of deleted bridges and add new bridges. */
571 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
572 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
579 SHASH_FOR_EACH (node, &new_br) {
580 const char *br_name = node->name;
581 const struct ovsrec_bridge *br_cfg = node->data;
582 br = shash_find_data(&old_br, br_name);
584 /* If the bridge datapath type has changed, we need to tear it
585 * down and recreate. */
586 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
588 bridge_create(br_cfg);
591 bridge_create(br_cfg);
594 shash_destroy(&old_br);
595 shash_destroy(&new_br);
597 /* Reconfigure all bridges. */
598 LIST_FOR_EACH (br, node, &all_bridges) {
599 bridge_reconfigure_one(br);
602 /* Add and delete ports on all datapaths.
604 * The kernel will reject any attempt to add a given port to a datapath if
605 * that port already belongs to a different datapath, so we must do all
606 * port deletions before any port additions. */
607 LIST_FOR_EACH (br, node, &all_bridges) {
608 struct dpif_port_dump dump;
609 struct shash want_ifaces;
610 struct dpif_port dpif_port;
612 bridge_get_all_ifaces(br, &want_ifaces);
613 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
614 if (!shash_find(&want_ifaces, dpif_port.name)
615 && strcmp(dpif_port.name, br->name)) {
616 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
618 VLOG_WARN("failed to remove %s interface from %s: %s",
619 dpif_port.name, dpif_name(br->dpif),
624 shash_destroy(&want_ifaces);
626 LIST_FOR_EACH (br, node, &all_bridges) {
627 struct shash cur_ifaces, want_ifaces;
628 struct dpif_port_dump dump;
629 struct dpif_port dpif_port;
631 /* Get the set of interfaces currently in this datapath. */
632 shash_init(&cur_ifaces);
633 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
634 struct dpif_port *port_info = xmalloc(sizeof *port_info);
635 dpif_port_clone(port_info, &dpif_port);
636 shash_add(&cur_ifaces, dpif_port.name, port_info);
639 /* Get the set of interfaces we want on this datapath. */
640 bridge_get_all_ifaces(br, &want_ifaces);
642 hmap_clear(&br->ifaces);
643 SHASH_FOR_EACH (node, &want_ifaces) {
644 const char *if_name = node->name;
645 struct iface *iface = node->data;
646 struct dpif_port *dpif_port;
650 type = iface ? iface->type : "internal";
651 dpif_port = shash_find_data(&cur_ifaces, if_name);
653 /* If we have a port or a netdev already, and it's not the type we
654 * want, then delete the port (if any) and close the netdev (if
656 if ((dpif_port && strcmp(dpif_port->type, type))
657 || (iface && iface->netdev
658 && strcmp(type, netdev_get_type(iface->netdev)))) {
660 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
667 netdev_close(iface->netdev);
668 iface->netdev = NULL;
672 /* If the port doesn't exist or we don't have the netdev open,
673 * we need to do more work. */
674 if (!dpif_port || (iface && !iface->netdev)) {
675 struct netdev_options options;
676 struct netdev *netdev;
679 /* First open the network device. */
680 options.name = if_name;
682 options.args = &args;
683 options.ethertype = NETDEV_ETH_TYPE_NONE;
687 shash_from_ovs_idl_map(iface->cfg->key_options,
688 iface->cfg->value_options,
689 iface->cfg->n_options, &args);
691 error = netdev_open(&options, &netdev);
692 shash_destroy(&args);
695 VLOG_WARN("could not open network device %s (%s)",
696 if_name, strerror(error));
700 /* Then add the port if we haven't already. */
702 error = dpif_port_add(br->dpif, netdev, NULL);
704 netdev_close(netdev);
705 if (error == EFBIG) {
706 VLOG_ERR("ran out of valid port numbers on %s",
707 dpif_name(br->dpif));
710 VLOG_WARN("failed to add %s interface to %s: %s",
711 if_name, dpif_name(br->dpif),
718 /* Update 'iface'. */
720 iface->netdev = netdev;
721 iface->enabled = iface_get_carrier(iface);
722 iface->up = iface->enabled;
724 } else if (iface && iface->netdev) {
728 shash_from_ovs_idl_map(iface->cfg->key_options,
729 iface->cfg->value_options,
730 iface->cfg->n_options, &args);
731 netdev_set_config(iface->netdev, &args);
732 shash_destroy(&args);
735 shash_destroy(&want_ifaces);
737 SHASH_FOR_EACH (node, &cur_ifaces) {
738 struct dpif_port *port_info = node->data;
739 dpif_port_destroy(port_info);
742 shash_destroy(&cur_ifaces);
744 sflow_bridge_number = 0;
745 LIST_FOR_EACH (br, node, &all_bridges) {
748 struct iface *local_iface;
749 struct iface *hw_addr_iface;
752 bridge_fetch_dp_ifaces(br);
754 iterate_and_prune_ifaces(br, check_iface, NULL);
756 /* Pick local port hardware address, datapath ID. */
757 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
758 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
760 int error = netdev_set_etheraddr(local_iface->netdev, ea);
762 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
763 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
764 "Ethernet address: %s",
765 br->name, strerror(error));
768 memcpy(br->ea, ea, ETH_ADDR_LEN);
770 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
771 ofproto_set_datapath_id(br->ofproto, dpid);
773 dpid_string = xasprintf("%016"PRIx64, dpid);
774 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
777 /* Set NetFlow configuration on this bridge. */
778 if (br->cfg->netflow) {
779 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
780 struct netflow_options opts;
782 memset(&opts, 0, sizeof opts);
784 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
785 if (nf_cfg->engine_type) {
786 opts.engine_type = *nf_cfg->engine_type;
788 if (nf_cfg->engine_id) {
789 opts.engine_id = *nf_cfg->engine_id;
792 opts.active_timeout = nf_cfg->active_timeout;
793 if (!opts.active_timeout) {
794 opts.active_timeout = -1;
795 } else if (opts.active_timeout < 0) {
796 VLOG_WARN("bridge %s: active timeout interval set to negative "
797 "value, using default instead (%d seconds)", br->name,
798 NF_ACTIVE_TIMEOUT_DEFAULT);
799 opts.active_timeout = -1;
802 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
803 if (opts.add_id_to_iface) {
804 if (opts.engine_id > 0x7f) {
805 VLOG_WARN("bridge %s: netflow port mangling may conflict "
806 "with another vswitch, choose an engine id less "
807 "than 128", br->name);
809 if (hmap_count(&br->ports) > 508) {
810 VLOG_WARN("bridge %s: netflow port mangling will conflict "
811 "with another port when more than 508 ports are "
816 opts.collectors.n = nf_cfg->n_targets;
817 opts.collectors.names = nf_cfg->targets;
818 if (ofproto_set_netflow(br->ofproto, &opts)) {
819 VLOG_ERR("bridge %s: problem setting netflow collectors",
823 ofproto_set_netflow(br->ofproto, NULL);
826 /* Set sFlow configuration on this bridge. */
827 if (br->cfg->sflow) {
828 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
829 struct ovsrec_controller **controllers;
830 struct ofproto_sflow_options oso;
831 size_t n_controllers;
833 memset(&oso, 0, sizeof oso);
835 oso.targets.n = sflow_cfg->n_targets;
836 oso.targets.names = sflow_cfg->targets;
838 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
839 if (sflow_cfg->sampling) {
840 oso.sampling_rate = *sflow_cfg->sampling;
843 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
844 if (sflow_cfg->polling) {
845 oso.polling_interval = *sflow_cfg->polling;
848 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
849 if (sflow_cfg->header) {
850 oso.header_len = *sflow_cfg->header;
853 oso.sub_id = sflow_bridge_number++;
854 oso.agent_device = sflow_cfg->agent;
856 oso.control_ip = NULL;
857 n_controllers = bridge_get_controllers(br, &controllers);
858 for (i = 0; i < n_controllers; i++) {
859 if (controllers[i]->local_ip) {
860 oso.control_ip = controllers[i]->local_ip;
864 ofproto_set_sflow(br->ofproto, &oso);
866 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
868 ofproto_set_sflow(br->ofproto, NULL);
871 /* Update the controller and related settings. It would be more
872 * straightforward to call this from bridge_reconfigure_one(), but we
873 * can't do it there for two reasons. First, and most importantly, at
874 * that point we don't know the dp_ifidx of any interfaces that have
875 * been added to the bridge (because we haven't actually added them to
876 * the datapath). Second, at that point we haven't set the datapath ID
877 * yet; when a controller is configured, resetting the datapath ID will
878 * immediately disconnect from the controller, so it's better to set
879 * the datapath ID before the controller. */
880 bridge_reconfigure_remotes(br, managers, n_managers);
882 LIST_FOR_EACH (br, node, &all_bridges) {
885 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
889 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
890 netdev_monitor_add(port->monitor, iface->netdev);
893 port->miimon_next_update = 0;
896 port_update_lacp(port);
897 port_update_bonding(port);
899 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
900 iface_update_qos(iface, port->cfg->qos);
904 LIST_FOR_EACH (br, node, &all_bridges) {
905 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
908 LIST_FOR_EACH (br, node, &all_bridges) {
910 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
911 iface_update_cfm(iface);
917 /* ovs-vswitchd has completed initialization, so allow the process that
918 * forked us to exit successfully. */
919 daemonize_complete();
923 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
924 const struct ovsdb_idl_column *column,
927 const struct ovsdb_datum *datum;
928 union ovsdb_atom atom;
931 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
932 atom.string = (char *) key;
933 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
934 return idx == UINT_MAX ? NULL : datum->values[idx].string;
938 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
940 return get_ovsrec_key_value(&br_cfg->header_,
941 &ovsrec_bridge_col_other_config, key);
945 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
946 struct iface **hw_addr_iface)
952 *hw_addr_iface = NULL;
954 /* Did the user request a particular MAC? */
955 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
956 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
957 if (eth_addr_is_multicast(ea)) {
958 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
959 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
960 } else if (eth_addr_is_zero(ea)) {
961 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
967 /* Otherwise choose the minimum non-local MAC address among all of the
969 memset(ea, 0xff, ETH_ADDR_LEN);
970 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
971 uint8_t iface_ea[ETH_ADDR_LEN];
972 struct iface *candidate;
975 /* Mirror output ports don't participate. */
976 if (port->is_mirror_output_port) {
980 /* Choose the MAC address to represent the port. */
982 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
983 /* Find the interface with this Ethernet address (if any) so that
984 * we can provide the correct devname to the caller. */
985 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
986 uint8_t candidate_ea[ETH_ADDR_LEN];
987 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
988 && eth_addr_equals(iface_ea, candidate_ea)) {
993 /* Choose the interface whose MAC address will represent the port.
994 * The Linux kernel bonding code always chooses the MAC address of
995 * the first slave added to a bond, and the Fedora networking
996 * scripts always add slaves to a bond in alphabetical order, so
997 * for compatibility we choose the interface with the name that is
998 * first in alphabetical order. */
999 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1000 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1005 /* The local port doesn't count (since we're trying to choose its
1006 * MAC address anyway). */
1007 if (iface->dp_ifidx == ODPP_LOCAL) {
1012 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1014 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1015 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1016 iface->name, strerror(error));
1021 /* Compare against our current choice. */
1022 if (!eth_addr_is_multicast(iface_ea) &&
1023 !eth_addr_is_local(iface_ea) &&
1024 !eth_addr_is_reserved(iface_ea) &&
1025 !eth_addr_is_zero(iface_ea) &&
1026 eth_addr_compare_3way(iface_ea, ea) < 0)
1028 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1029 *hw_addr_iface = iface;
1032 if (eth_addr_is_multicast(ea)) {
1033 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1034 *hw_addr_iface = NULL;
1035 VLOG_WARN("bridge %s: using default bridge Ethernet "
1036 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1038 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1039 br->name, ETH_ADDR_ARGS(ea));
1043 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1044 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1045 * an interface on 'br', then that interface must be passed in as
1046 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1047 * 'hw_addr_iface' must be passed in as a null pointer. */
1049 bridge_pick_datapath_id(struct bridge *br,
1050 const uint8_t bridge_ea[ETH_ADDR_LEN],
1051 struct iface *hw_addr_iface)
1054 * The procedure for choosing a bridge MAC address will, in the most
1055 * ordinary case, also choose a unique MAC that we can use as a datapath
1056 * ID. In some special cases, though, multiple bridges will end up with
1057 * the same MAC address. This is OK for the bridges, but it will confuse
1058 * the OpenFlow controller, because each datapath needs a unique datapath
1061 * Datapath IDs must be unique. It is also very desirable that they be
1062 * stable from one run to the next, so that policy set on a datapath
1065 const char *datapath_id;
1068 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1069 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1073 if (hw_addr_iface) {
1075 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1077 * A bridge whose MAC address is taken from a VLAN network device
1078 * (that is, a network device created with vconfig(8) or similar
1079 * tool) will have the same MAC address as a bridge on the VLAN
1080 * device's physical network device.
1082 * Handle this case by hashing the physical network device MAC
1083 * along with the VLAN identifier.
1085 uint8_t buf[ETH_ADDR_LEN + 2];
1086 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1087 buf[ETH_ADDR_LEN] = vlan >> 8;
1088 buf[ETH_ADDR_LEN + 1] = vlan;
1089 return dpid_from_hash(buf, sizeof buf);
1092 * Assume that this bridge's MAC address is unique, since it
1093 * doesn't fit any of the cases we handle specially.
1098 * A purely internal bridge, that is, one that has no non-virtual
1099 * network devices on it at all, is more difficult because it has no
1100 * natural unique identifier at all.
1102 * When the host is a XenServer, we handle this case by hashing the
1103 * host's UUID with the name of the bridge. Names of bridges are
1104 * persistent across XenServer reboots, although they can be reused if
1105 * an internal network is destroyed and then a new one is later
1106 * created, so this is fairly effective.
1108 * When the host is not a XenServer, we punt by using a random MAC
1109 * address on each run.
1111 const char *host_uuid = xenserver_get_host_uuid();
1113 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1114 dpid = dpid_from_hash(combined, strlen(combined));
1120 return eth_addr_to_uint64(bridge_ea);
1124 dpid_from_hash(const void *data, size_t n)
1126 uint8_t hash[SHA1_DIGEST_SIZE];
1128 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1129 sha1_bytes(data, n, hash);
1130 eth_addr_mark_random(hash);
1131 return eth_addr_to_uint64(hash);
1135 iface_refresh_status(struct iface *iface)
1139 enum netdev_flags flags;
1148 if (!netdev_get_status(iface->netdev, &sh)) {
1150 char **keys, **values;
1152 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1153 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1158 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1161 shash_destroy_free_data(&sh);
1163 error = netdev_get_flags(iface->netdev, &flags);
1165 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1168 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1171 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1173 ovsrec_interface_set_duplex(iface->cfg,
1174 netdev_features_is_full_duplex(current)
1176 /* warning: uint64_t -> int64_t conversion */
1177 bps = netdev_features_to_bps(current);
1178 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1181 ovsrec_interface_set_duplex(iface->cfg, NULL);
1182 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1186 ovsrec_interface_set_link_state(iface->cfg,
1187 iface_get_carrier(iface) ? "up" : "down");
1189 error = netdev_get_mtu(iface->netdev, &mtu);
1190 if (!error && mtu != INT_MAX) {
1192 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1195 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1200 iface_refresh_cfm_stats(struct iface *iface)
1202 const struct ovsrec_monitor *mon;
1203 const struct cfm *cfm;
1206 mon = iface->cfg->monitor;
1207 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1213 for (i = 0; i < mon->n_remote_mps; i++) {
1214 const struct ovsrec_maintenance_point *mp;
1215 const struct remote_mp *rmp;
1217 mp = mon->remote_mps[i];
1218 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1220 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1223 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1227 iface_refresh_stats(struct iface *iface)
1233 static const struct iface_stat iface_stats[] = {
1234 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1235 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1236 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1237 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1238 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1239 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1240 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1241 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1242 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1243 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1244 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1245 { "collisions", offsetof(struct netdev_stats, collisions) },
1247 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1248 const struct iface_stat *s;
1250 char *keys[N_STATS];
1251 int64_t values[N_STATS];
1254 struct netdev_stats stats;
1256 /* Intentionally ignore return value, since errors will set 'stats' to
1257 * all-1s, and we will deal with that correctly below. */
1258 netdev_get_stats(iface->netdev, &stats);
1261 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1262 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1263 if (value != UINT64_MAX) {
1270 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1274 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1276 struct ovsdb_datum datum;
1280 get_system_stats(&stats);
1282 ovsdb_datum_from_shash(&datum, &stats);
1283 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1287 static inline const char *
1288 nx_role_to_str(enum nx_role role)
1293 case NX_ROLE_MASTER:
1298 return "*** INVALID ROLE ***";
1303 bridge_refresh_controller_status(const struct bridge *br)
1306 const struct ovsrec_controller *cfg;
1308 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1310 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1311 struct ofproto_controller_info *cinfo =
1312 shash_find_data(&info, cfg->target);
1315 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1316 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1317 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1318 (char **) cinfo->pairs.values,
1321 ovsrec_controller_set_is_connected(cfg, false);
1322 ovsrec_controller_set_role(cfg, NULL);
1323 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1327 ofproto_free_ofproto_controller_info(&info);
1333 const struct ovsrec_open_vswitch *cfg;
1335 bool datapath_destroyed;
1336 bool database_changed;
1339 /* Let each bridge do the work that it needs to do. */
1340 datapath_destroyed = false;
1341 LIST_FOR_EACH (br, node, &all_bridges) {
1342 int error = bridge_run_one(br);
1344 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1345 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1346 "forcing reconfiguration", br->name);
1347 datapath_destroyed = true;
1351 /* (Re)configure if necessary. */
1352 database_changed = ovsdb_idl_run(idl);
1353 cfg = ovsrec_open_vswitch_first(idl);
1355 /* Re-configure SSL. We do this on every trip through the main loop,
1356 * instead of just when the database changes, because the contents of the
1357 * key and certificate files can change without the database changing.
1359 * We do this before bridge_reconfigure() because that function might
1360 * initiate SSL connections and thus requires SSL to be configured. */
1361 if (cfg && cfg->ssl) {
1362 const struct ovsrec_ssl *ssl = cfg->ssl;
1364 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1365 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1368 if (database_changed || datapath_destroyed) {
1370 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1372 bridge_configure_once(cfg);
1373 bridge_reconfigure(cfg);
1375 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1376 ovsdb_idl_txn_commit(txn);
1377 ovsdb_idl_txn_destroy(txn); /* XXX */
1379 /* We still need to reconfigure to avoid dangling pointers to
1380 * now-destroyed ovsrec structures inside bridge data. */
1381 static const struct ovsrec_open_vswitch null_cfg;
1383 bridge_reconfigure(&null_cfg);
1387 /* Refresh system and interface stats if necessary. */
1388 if (time_msec() >= stats_timer) {
1390 struct ovsdb_idl_txn *txn;
1392 txn = ovsdb_idl_txn_create(idl);
1393 LIST_FOR_EACH (br, node, &all_bridges) {
1396 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1397 struct iface *iface;
1399 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1400 iface_refresh_stats(iface);
1401 iface_refresh_cfm_stats(iface);
1402 iface_refresh_status(iface);
1405 bridge_refresh_controller_status(br);
1407 refresh_system_stats(cfg);
1408 ovsdb_idl_txn_commit(txn);
1409 ovsdb_idl_txn_destroy(txn); /* XXX */
1412 stats_timer = time_msec() + STATS_INTERVAL;
1421 LIST_FOR_EACH (br, node, &all_bridges) {
1424 ofproto_wait(br->ofproto);
1425 if (ofproto_has_primary_controller(br->ofproto)) {
1429 mac_learning_wait(br->ml);
1431 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1435 ovsdb_idl_wait(idl);
1436 poll_timer_wait_until(stats_timer);
1439 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1440 * configuration changes. */
1442 bridge_flush(struct bridge *br)
1444 COVERAGE_INC(bridge_flush);
1448 /* Bridge unixctl user interface functions. */
1450 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1451 const char *args, void *aux OVS_UNUSED)
1453 struct ds ds = DS_EMPTY_INITIALIZER;
1454 const struct bridge *br;
1455 const struct mac_entry *e;
1457 br = bridge_lookup(args);
1459 unixctl_command_reply(conn, 501, "no such bridge");
1463 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1464 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1465 struct port *port = e->port.p;
1466 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1467 port_get_an_iface(port)->dp_ifidx,
1468 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1470 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1474 /* QoS unixctl user interface functions. */
1476 struct qos_unixctl_show_cbdata {
1478 struct iface *iface;
1482 qos_unixctl_show_cb(unsigned int queue_id,
1483 const struct shash *details,
1486 struct qos_unixctl_show_cbdata *data = aux;
1487 struct ds *ds = data->ds;
1488 struct iface *iface = data->iface;
1489 struct netdev_queue_stats stats;
1490 struct shash_node *node;
1493 ds_put_cstr(ds, "\n");
1495 ds_put_format(ds, "Queue %u:\n", queue_id);
1497 ds_put_cstr(ds, "Default:\n");
1500 SHASH_FOR_EACH (node, details) {
1501 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1504 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1506 if (stats.tx_packets != UINT64_MAX) {
1507 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1510 if (stats.tx_bytes != UINT64_MAX) {
1511 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1514 if (stats.tx_errors != UINT64_MAX) {
1515 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1518 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1519 queue_id, strerror(error));
1524 qos_unixctl_show(struct unixctl_conn *conn,
1525 const char *args, void *aux OVS_UNUSED)
1527 struct ds ds = DS_EMPTY_INITIALIZER;
1528 struct shash sh = SHASH_INITIALIZER(&sh);
1529 struct iface *iface;
1531 struct shash_node *node;
1532 struct qos_unixctl_show_cbdata data;
1535 iface = iface_find(args);
1537 unixctl_command_reply(conn, 501, "no such interface");
1541 netdev_get_qos(iface->netdev, &type, &sh);
1543 if (*type != '\0') {
1544 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1546 SHASH_FOR_EACH (node, &sh) {
1547 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1552 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1555 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1557 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1559 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1560 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1563 shash_destroy_free_data(&sh);
1567 /* Bridge reconfiguration functions. */
1568 static struct bridge *
1569 bridge_create(const struct ovsrec_bridge *br_cfg)
1574 assert(!bridge_lookup(br_cfg->name));
1575 br = xzalloc(sizeof *br);
1577 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1583 dpif_flow_flush(br->dpif);
1585 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1588 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1590 dpif_delete(br->dpif);
1591 dpif_close(br->dpif);
1596 br->name = xstrdup(br_cfg->name);
1598 br->ml = mac_learning_create();
1599 eth_addr_nicira_random(br->default_ea);
1601 hmap_init(&br->ports);
1602 hmap_init(&br->ifaces);
1603 shash_init(&br->iface_by_name);
1607 list_push_back(&all_bridges, &br->node);
1609 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1615 bridge_destroy(struct bridge *br)
1618 struct port *port, *next;
1621 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1624 list_remove(&br->node);
1625 error = dpif_delete(br->dpif);
1626 if (error && error != ENOENT) {
1627 VLOG_ERR("failed to delete %s: %s",
1628 dpif_name(br->dpif), strerror(error));
1630 dpif_close(br->dpif);
1631 ofproto_destroy(br->ofproto);
1632 mac_learning_destroy(br->ml);
1633 hmap_destroy(&br->ifaces);
1634 hmap_destroy(&br->ports);
1635 shash_destroy(&br->iface_by_name);
1641 static struct bridge *
1642 bridge_lookup(const char *name)
1646 LIST_FOR_EACH (br, node, &all_bridges) {
1647 if (!strcmp(br->name, name)) {
1654 /* Handle requests for a listing of all flows known by the OpenFlow
1655 * stack, including those normally hidden. */
1657 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1658 const char *args, void *aux OVS_UNUSED)
1663 br = bridge_lookup(args);
1665 unixctl_command_reply(conn, 501, "Unknown bridge");
1670 ofproto_get_all_flows(br->ofproto, &results);
1672 unixctl_command_reply(conn, 200, ds_cstr(&results));
1673 ds_destroy(&results);
1676 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1677 * connections and reconnect. If BRIDGE is not specified, then all bridges
1678 * drop their controller connections and reconnect. */
1680 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1681 const char *args, void *aux OVS_UNUSED)
1684 if (args[0] != '\0') {
1685 br = bridge_lookup(args);
1687 unixctl_command_reply(conn, 501, "Unknown bridge");
1690 ofproto_reconnect_controllers(br->ofproto);
1692 LIST_FOR_EACH (br, node, &all_bridges) {
1693 ofproto_reconnect_controllers(br->ofproto);
1696 unixctl_command_reply(conn, 200, NULL);
1700 bridge_run_one(struct bridge *br)
1705 error = ofproto_run1(br->ofproto);
1710 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1712 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1716 error = ofproto_run2(br->ofproto, br->flush);
1723 bridge_get_controllers(const struct bridge *br,
1724 struct ovsrec_controller ***controllersp)
1726 struct ovsrec_controller **controllers;
1727 size_t n_controllers;
1729 controllers = br->cfg->controller;
1730 n_controllers = br->cfg->n_controller;
1732 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1738 *controllersp = controllers;
1740 return n_controllers;
1744 bridge_reconfigure_one(struct bridge *br)
1746 enum ofproto_fail_mode fail_mode;
1747 struct svec snoops, old_snoops;
1748 struct port *port, *next;
1749 struct shash_node *node;
1750 struct shash new_ports;
1753 /* Collect new ports. */
1754 shash_init(&new_ports);
1755 for (i = 0; i < br->cfg->n_ports; i++) {
1756 const char *name = br->cfg->ports[i]->name;
1757 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1758 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1763 /* If we have a controller, then we need a local port. Complain if the
1764 * user didn't specify one.
1766 * XXX perhaps we should synthesize a port ourselves in this case. */
1767 if (bridge_get_controllers(br, NULL)) {
1768 char local_name[IF_NAMESIZE];
1771 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1772 local_name, sizeof local_name);
1773 if (!error && !shash_find(&new_ports, local_name)) {
1774 VLOG_WARN("bridge %s: controller specified but no local port "
1775 "(port named %s) defined",
1776 br->name, local_name);
1780 /* Get rid of deleted ports.
1781 * Get rid of deleted interfaces on ports that still exist. */
1782 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1783 const struct ovsrec_port *port_cfg;
1785 port_cfg = shash_find_data(&new_ports, port->name);
1789 port_del_ifaces(port, port_cfg);
1793 /* Create new ports.
1794 * Add new interfaces to existing ports.
1795 * Reconfigure existing ports. */
1796 SHASH_FOR_EACH (node, &new_ports) {
1797 struct port *port = port_lookup(br, node->name);
1799 port = port_create(br, node->name);
1802 port_reconfigure(port, node->data);
1803 if (!port->n_ifaces) {
1804 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1805 br->name, port->name);
1809 shash_destroy(&new_ports);
1811 /* Set the fail-mode */
1812 fail_mode = !br->cfg->fail_mode
1813 || !strcmp(br->cfg->fail_mode, "standalone")
1814 ? OFPROTO_FAIL_STANDALONE
1815 : OFPROTO_FAIL_SECURE;
1816 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1817 && !ofproto_has_primary_controller(br->ofproto)) {
1818 ofproto_flush_flows(br->ofproto);
1820 ofproto_set_fail_mode(br->ofproto, fail_mode);
1822 /* Delete all flows if we're switching from connected to standalone or vice
1823 * versa. (XXX Should we delete all flows if we are switching from one
1824 * controller to another?) */
1826 /* Configure OpenFlow controller connection snooping. */
1828 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1829 ovs_rundir(), br->name));
1830 svec_init(&old_snoops);
1831 ofproto_get_snoops(br->ofproto, &old_snoops);
1832 if (!svec_equal(&snoops, &old_snoops)) {
1833 ofproto_set_snoops(br->ofproto, &snoops);
1835 svec_destroy(&snoops);
1836 svec_destroy(&old_snoops);
1838 mirror_reconfigure(br);
1841 /* Initializes 'oc' appropriately as a management service controller for
1844 * The caller must free oc->target when it is no longer needed. */
1846 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1847 struct ofproto_controller *oc)
1849 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1850 oc->max_backoff = 0;
1851 oc->probe_interval = 60;
1852 oc->band = OFPROTO_OUT_OF_BAND;
1854 oc->burst_limit = 0;
1857 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1859 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1860 struct ofproto_controller *oc)
1862 oc->target = c->target;
1863 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1864 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1865 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1866 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1867 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1868 oc->burst_limit = (c->controller_burst_limit
1869 ? *c->controller_burst_limit : 0);
1872 /* Configures the IP stack for 'br''s local interface properly according to the
1873 * configuration in 'c'. */
1875 bridge_configure_local_iface_netdev(struct bridge *br,
1876 struct ovsrec_controller *c)
1878 struct netdev *netdev;
1879 struct in_addr mask, gateway;
1881 struct iface *local_iface;
1884 /* If there's no local interface or no IP address, give up. */
1885 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1886 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1890 /* Bring up the local interface. */
1891 netdev = local_iface->netdev;
1892 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1894 /* Configure the IP address and netmask. */
1895 if (!c->local_netmask
1896 || !inet_aton(c->local_netmask, &mask)
1898 mask.s_addr = guess_netmask(ip.s_addr);
1900 if (!netdev_set_in4(netdev, ip, mask)) {
1901 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1902 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1905 /* Configure the default gateway. */
1906 if (c->local_gateway
1907 && inet_aton(c->local_gateway, &gateway)
1908 && gateway.s_addr) {
1909 if (!netdev_add_router(netdev, gateway)) {
1910 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1911 br->name, IP_ARGS(&gateway.s_addr));
1917 bridge_reconfigure_remotes(struct bridge *br,
1918 const struct sockaddr_in *managers,
1921 const char *disable_ib_str, *queue_id_str;
1922 bool disable_in_band = false;
1925 struct ovsrec_controller **controllers;
1926 size_t n_controllers;
1929 struct ofproto_controller *ocs;
1933 /* Check if we should disable in-band control on this bridge. */
1934 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1935 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1936 disable_in_band = true;
1939 /* Set OpenFlow queue ID for in-band control. */
1940 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1941 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1942 ofproto_set_in_band_queue(br->ofproto, queue_id);
1944 if (disable_in_band) {
1945 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1947 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1949 had_primary = ofproto_has_primary_controller(br->ofproto);
1951 n_controllers = bridge_get_controllers(br, &controllers);
1953 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1956 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1957 for (i = 0; i < n_controllers; i++) {
1958 struct ovsrec_controller *c = controllers[i];
1960 if (!strncmp(c->target, "punix:", 6)
1961 || !strncmp(c->target, "unix:", 5)) {
1962 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1964 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1965 * domain sockets and overwriting arbitrary local files. */
1966 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1967 "\"%s\" due to possibility for remote exploit",
1968 dpif_name(br->dpif), c->target);
1972 bridge_configure_local_iface_netdev(br, c);
1973 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1974 if (disable_in_band) {
1975 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1980 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1981 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1984 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1985 ofproto_flush_flows(br->ofproto);
1988 /* If there are no controllers and the bridge is in standalone
1989 * mode, set up a flow that matches every packet and directs
1990 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1991 * switch is in secure mode and we won't pass any traffic until
1992 * a controller has been defined and it tells us to do so. */
1994 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1995 union ofp_action action;
1996 struct cls_rule rule;
1998 memset(&action, 0, sizeof action);
1999 action.type = htons(OFPAT_OUTPUT);
2000 action.output.len = htons(sizeof action);
2001 action.output.port = htons(OFPP_NORMAL);
2002 cls_rule_init_catchall(&rule, 0);
2003 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2008 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2013 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2014 struct iface *iface;
2016 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2017 shash_add_once(ifaces, iface->name, iface);
2019 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2020 shash_add_once(ifaces, port->name, NULL);
2025 /* For robustness, in case the administrator moves around datapath ports behind
2026 * our back, we re-check all the datapath port numbers here.
2028 * This function will set the 'dp_ifidx' members of interfaces that have
2029 * disappeared to -1, so only call this function from a context where those
2030 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2031 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2032 * datapath, which doesn't support UINT16_MAX+1 ports. */
2034 bridge_fetch_dp_ifaces(struct bridge *br)
2036 struct dpif_port_dump dump;
2037 struct dpif_port dpif_port;
2040 /* Reset all interface numbers. */
2041 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2042 struct iface *iface;
2044 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2045 iface->dp_ifidx = -1;
2048 hmap_clear(&br->ifaces);
2050 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2051 struct iface *iface = iface_lookup(br, dpif_port.name);
2053 if (iface->dp_ifidx >= 0) {
2054 VLOG_WARN("%s reported interface %s twice",
2055 dpif_name(br->dpif), dpif_port.name);
2056 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2057 VLOG_WARN("%s reported interface %"PRIu16" twice",
2058 dpif_name(br->dpif), dpif_port.port_no);
2060 iface->dp_ifidx = dpif_port.port_no;
2061 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2062 hash_int(iface->dp_ifidx, 0));
2065 iface_set_ofport(iface->cfg,
2066 (iface->dp_ifidx >= 0
2067 ? odp_port_to_ofp_port(iface->dp_ifidx)
2073 /* Bridge packet processing functions. */
2076 bond_is_tcp_hash(const struct port *port)
2078 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2082 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2084 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2087 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2089 struct flow hash_flow;
2091 memcpy(&hash_flow, flow, sizeof hash_flow);
2092 hash_flow.vlan_tci = 0;
2094 /* The symmetric quality of this hash function is not required, but
2095 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2096 * purposes, so we use it out of convenience. */
2097 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2100 static struct bond_entry *
2101 lookup_bond_entry(const struct port *port, const struct flow *flow,
2104 assert(port->bond_mode != BM_AB);
2106 if (bond_is_tcp_hash(port)) {
2107 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2109 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2113 static struct iface *
2114 bond_choose_iface(const struct port *port)
2116 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2117 struct iface *best_down_slave;
2118 struct iface *iface;
2120 best_down_slave = NULL;
2121 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2122 if (iface->enabled) {
2124 } else if ((!best_down_slave
2125 || iface->delay_expires < best_down_slave->delay_expires)
2126 && lacp_slave_may_enable(port->lacp, iface)) {
2127 best_down_slave = iface;
2131 if (best_down_slave) {
2132 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2133 "since no other interface is up",
2134 best_down_slave->name,
2135 best_down_slave->delay_expires - time_msec());
2136 bond_enable_slave(best_down_slave, true);
2139 return best_down_slave;
2143 choose_output_iface(const struct port *port, const struct flow *flow,
2144 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2146 struct iface *iface;
2148 assert(port->n_ifaces);
2149 if (port->n_ifaces == 1) {
2150 iface = port_get_an_iface(port);
2151 } else if (port->bond_mode == BM_AB) {
2152 iface = port->active_iface;
2154 *tags |= port->no_ifaces_tag;
2158 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2159 if (!e->iface || !e->iface->enabled) {
2160 /* XXX select interface properly. The current interface selection
2161 * is only good for testing the rebalancing code. */
2162 e->iface = bond_choose_iface(port);
2164 *tags |= port->no_ifaces_tag;
2167 e->tag = tag_create_random();
2172 *dp_ifidx = iface->dp_ifidx;
2173 *tags |= iface->tag; /* Currently only used for bonding. */
2178 bond_link_status_update(struct iface *iface)
2180 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2181 struct port *port = iface->port;
2182 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2183 int updelay, downdelay;
2185 updelay = port->updelay;
2186 downdelay = port->downdelay;
2188 if (lacp_negotiated(port->lacp)) {
2193 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2194 /* Nothing to do. */
2197 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2198 iface->name, up ? "up" : "down");
2199 if (up == iface->enabled) {
2200 iface->delay_expires = LLONG_MAX;
2201 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2202 iface->name, up ? "disabled" : "enabled");
2203 } else if (up && !port->active_iface) {
2204 bond_enable_slave(iface, true);
2206 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2207 "other interface is up", iface->name, updelay);
2210 int delay = up ? updelay : downdelay;
2211 iface->delay_expires = time_msec() + delay;
2214 "interface %s: will be %s if it stays %s for %d ms",
2216 up ? "enabled" : "disabled",
2224 bond_choose_active_iface(struct port *port)
2226 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2228 port->active_iface = bond_choose_iface(port);
2229 if (port->active_iface) {
2230 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2231 port->name, port->active_iface->name);
2233 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2239 bond_enable_slave(struct iface *iface, bool enable)
2241 struct port *port = iface->port;
2242 struct bridge *br = port->bridge;
2244 /* This acts as a recursion check. If the act of disabling a slave
2245 * causes a different slave to be enabled, the flag will allow us to
2246 * skip redundant work when we reenter this function. It must be
2247 * cleared on exit to keep things safe with multiple bonds. */
2248 static bool moving_active_iface = false;
2250 iface->delay_expires = LLONG_MAX;
2251 if (enable == iface->enabled) {
2255 iface->enabled = enable;
2256 if (!iface->enabled) {
2257 VLOG_WARN("interface %s: disabled", iface->name);
2258 ofproto_revalidate(br->ofproto, iface->tag);
2259 if (iface == port->active_iface) {
2260 /* Disabling a slave can lead to another slave being immediately
2261 * enabled if there will be no active slaves but one is waiting
2262 * on an updelay. In this case we do not need to run most of the
2263 * code for the newly enabled slave since there was no period
2264 * without an active slave and it is redundant with the disabling
2266 moving_active_iface = true;
2267 bond_choose_active_iface(port);
2269 bond_send_learning_packets(port);
2271 VLOG_WARN("interface %s: enabled", iface->name);
2272 if (!port->active_iface && !moving_active_iface) {
2273 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2274 bond_choose_active_iface(port);
2275 bond_send_learning_packets(port);
2277 iface->tag = tag_create_random();
2280 moving_active_iface = false;
2283 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2284 * bond interface. */
2286 bond_update_fake_iface_stats(struct port *port)
2288 struct netdev_stats bond_stats;
2289 struct netdev *bond_dev;
2290 struct iface *iface;
2292 memset(&bond_stats, 0, sizeof bond_stats);
2294 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2295 struct netdev_stats slave_stats;
2297 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2298 /* XXX: We swap the stats here because they are swapped back when
2299 * reported by the internal device. The reason for this is
2300 * internal devices normally represent packets going into the system
2301 * but when used as fake bond device they represent packets leaving
2302 * the system. We really should do this in the internal device
2303 * itself because changing it here reverses the counts from the
2304 * perspective of the switch. However, the internal device doesn't
2305 * know what type of device it represents so we have to do it here
2307 bond_stats.tx_packets += slave_stats.rx_packets;
2308 bond_stats.tx_bytes += slave_stats.rx_bytes;
2309 bond_stats.rx_packets += slave_stats.tx_packets;
2310 bond_stats.rx_bytes += slave_stats.tx_bytes;
2314 if (!netdev_open_default(port->name, &bond_dev)) {
2315 netdev_set_stats(bond_dev, &bond_stats);
2316 netdev_close(bond_dev);
2321 bond_run(struct port *port)
2323 struct iface *iface;
2325 if (port->n_ifaces < 2) {
2329 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2330 bond_link_status_update(iface);
2333 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2334 if (time_msec() >= iface->delay_expires) {
2335 bond_enable_slave(iface, !iface->enabled);
2339 if (port->bond_fake_iface
2340 && time_msec() >= port->bond_next_fake_iface_update) {
2341 bond_update_fake_iface_stats(port);
2342 port->bond_next_fake_iface_update = time_msec() + 1000;
2347 bond_wait(struct port *port)
2349 struct iface *iface;
2351 if (port->n_ifaces < 2) {
2355 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2356 if (iface->delay_expires != LLONG_MAX) {
2357 poll_timer_wait_until(iface->delay_expires);
2361 if (port->bond_fake_iface) {
2362 poll_timer_wait_until(port->bond_next_fake_iface_update);
2367 set_dst(struct dst *dst, const struct flow *flow,
2368 const struct port *in_port, const struct port *out_port,
2371 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2372 : in_port->vlan >= 0 ? in_port->vlan
2373 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2374 : vlan_tci_to_vid(flow->vlan_tci));
2375 return choose_output_iface(out_port, flow, dst->vlan,
2376 &dst->dp_ifidx, tags);
2380 swap_dst(struct dst *p, struct dst *q)
2382 struct dst tmp = *p;
2387 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2388 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2389 * that we push to the datapath. We could in fact fully sort the array by
2390 * vlan, but in most cases there are at most two different vlan tags so that's
2391 * possibly overkill.) */
2393 partition_dsts(struct dst_set *set, int vlan)
2395 struct dst *first = set->dsts;
2396 struct dst *last = set->dsts + set->n;
2398 while (first != last) {
2400 * - All dsts < first have vlan == 'vlan'.
2401 * - All dsts >= last have vlan != 'vlan'.
2402 * - first < last. */
2403 while (first->vlan == vlan) {
2404 if (++first == last) {
2409 /* Same invariants, plus one additional:
2410 * - first->vlan != vlan.
2412 while (last[-1].vlan != vlan) {
2413 if (--last == first) {
2418 /* Same invariants, plus one additional:
2419 * - last[-1].vlan == vlan.*/
2420 swap_dst(first++, --last);
2425 mirror_mask_ffs(mirror_mask_t mask)
2427 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2432 dst_set_init(struct dst_set *set)
2434 set->dsts = set->builtin;
2436 set->allocated = ARRAY_SIZE(set->builtin);
2440 dst_set_add(struct dst_set *set, const struct dst *dst)
2442 if (set->n >= set->allocated) {
2443 size_t new_allocated;
2444 struct dst *new_dsts;
2446 new_allocated = set->allocated * 2;
2447 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2448 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2452 set->dsts = new_dsts;
2453 set->allocated = new_allocated;
2455 set->dsts[set->n++] = *dst;
2459 dst_set_free(struct dst_set *set)
2461 if (set->dsts != set->builtin) {
2467 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2470 for (i = 0; i < set->n; i++) {
2471 if (set->dsts[i].vlan == test->vlan
2472 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2480 port_trunks_vlan(const struct port *port, uint16_t vlan)
2482 return (port->vlan < 0
2483 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2487 port_includes_vlan(const struct port *port, uint16_t vlan)
2489 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2493 port_is_floodable(const struct port *port)
2495 struct iface *iface;
2497 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2498 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2506 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2507 * there is no active iface. */
2509 port_get_active_iface_tag(const struct port *port)
2511 return (port->active_iface
2512 ? port->active_iface->tag
2513 : port->no_ifaces_tag);
2516 /* Returns an arbitrary interface within 'port'.
2518 * 'port' must have at least one interface. */
2519 static struct iface *
2520 port_get_an_iface(const struct port *port)
2522 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2526 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2527 const struct port *in_port, const struct port *out_port,
2528 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2530 mirror_mask_t mirrors = in_port->src_mirrors;
2534 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2535 if (flow_vlan == 0) {
2536 flow_vlan = OFP_VLAN_NONE;
2539 if (out_port == FLOOD_PORT) {
2542 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2544 && port_is_floodable(port)
2545 && port_includes_vlan(port, vlan)
2546 && !port->is_mirror_output_port
2547 && set_dst(&dst, flow, in_port, port, tags)) {
2548 mirrors |= port->dst_mirrors;
2549 dst_set_add(set, &dst);
2552 *nf_output_iface = NF_OUT_FLOOD;
2553 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2554 dst_set_add(set, &dst);
2555 *nf_output_iface = dst.dp_ifidx;
2556 mirrors |= out_port->dst_mirrors;
2560 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2561 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2563 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2564 && !dst_is_duplicate(set, &dst)) {
2565 dst_set_add(set, &dst);
2570 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2571 if (port_includes_vlan(port, m->out_vlan)
2572 && set_dst(&dst, flow, in_port, port, tags))
2574 if (port->vlan < 0) {
2575 dst.vlan = m->out_vlan;
2577 if (dst_is_duplicate(set, &dst)) {
2581 /* Use the vlan tag on the original flow instead of
2582 * the one passed in the vlan parameter. This ensures
2583 * that we compare the vlan from before any implicit
2584 * tagging tags place. This is necessary because
2585 * dst->vlan is the final vlan, after removing implicit
2587 if (port == in_port && dst.vlan == flow_vlan) {
2588 /* Don't send out input port on same VLAN. */
2591 dst_set_add(set, &dst);
2596 mirrors &= mirrors - 1;
2599 partition_dsts(set, flow_vlan);
2602 static void OVS_UNUSED
2603 print_dsts(const struct dst_set *set)
2607 for (i = 0; i < set->n; i++) {
2608 const struct dst *dst = &set->dsts[i];
2610 printf(">p%"PRIu16, dst->dp_ifidx);
2611 if (dst->vlan != OFP_VLAN_NONE) {
2612 printf("v%"PRIu16, dst->vlan);
2618 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2619 const struct port *in_port, const struct port *out_port,
2620 tag_type *tags, struct ofpbuf *actions,
2621 uint16_t *nf_output_iface)
2628 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2631 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2632 if (cur_vlan == 0) {
2633 cur_vlan = OFP_VLAN_NONE;
2635 for (i = 0; i < set.n; i++) {
2636 const struct dst *dst = &set.dsts[i];
2637 if (dst->vlan != cur_vlan) {
2638 if (dst->vlan == OFP_VLAN_NONE) {
2639 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2642 tci = htons(dst->vlan & VLAN_VID_MASK);
2643 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2644 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2646 cur_vlan = dst->vlan;
2648 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2653 /* Returns the effective vlan of a packet, taking into account both the
2654 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2655 * the packet is untagged and -1 indicates it has an invalid header and
2656 * should be dropped. */
2657 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2658 struct port *in_port, bool have_packet)
2660 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2661 if (in_port->vlan >= 0) {
2663 /* XXX support double tagging? */
2665 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2666 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2667 "packet received on port %s configured with "
2668 "implicit VLAN %"PRIu16,
2669 br->name, vlan, in_port->name, in_port->vlan);
2673 vlan = in_port->vlan;
2675 if (!port_includes_vlan(in_port, vlan)) {
2677 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2678 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2679 "packet received on port %s not configured for "
2681 br->name, vlan, in_port->name, vlan);
2690 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2691 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2692 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2694 is_gratuitous_arp(const struct flow *flow)
2696 return (flow->dl_type == htons(ETH_TYPE_ARP)
2697 && eth_addr_is_broadcast(flow->dl_dst)
2698 && (flow->nw_proto == ARP_OP_REPLY
2699 || (flow->nw_proto == ARP_OP_REQUEST
2700 && flow->nw_src == flow->nw_dst)));
2704 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2705 struct port *in_port)
2707 struct mac_entry *mac;
2709 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2713 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2714 if (is_gratuitous_arp(flow)) {
2715 /* We don't want to learn from gratuitous ARP packets that are
2716 * reflected back over bond slaves so we lock the learning table. */
2717 if (in_port->n_ifaces == 1) {
2718 mac_entry_set_grat_arp_lock(mac);
2719 } else if (mac_entry_is_grat_arp_locked(mac)) {
2724 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2725 /* The log messages here could actually be useful in debugging,
2726 * so keep the rate limit relatively high. */
2727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2728 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2729 "on port %s in VLAN %d",
2730 br->name, ETH_ADDR_ARGS(flow->dl_src),
2731 in_port->name, vlan);
2733 mac->port.p = in_port;
2734 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2738 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2739 * dropped. Returns true if they may be forwarded, false if they should be
2742 * If 'have_packet' is true, it indicates that the caller is processing a
2743 * received packet. If 'have_packet' is false, then the caller is just
2744 * revalidating an existing flow because configuration has changed. Either
2745 * way, 'have_packet' only affects logging (there is no point in logging errors
2746 * during revalidation).
2748 * Sets '*in_portp' to the input port. This will be a null pointer if
2749 * flow->in_port does not designate a known input port (in which case
2750 * is_admissible() returns false).
2752 * When returning true, sets '*vlanp' to the effective VLAN of the input
2753 * packet, as returned by flow_get_vlan().
2755 * May also add tags to '*tags', although the current implementation only does
2756 * so in one special case.
2759 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2760 tag_type *tags, int *vlanp, struct port **in_portp)
2762 struct iface *in_iface;
2763 struct port *in_port;
2766 /* Find the interface and port structure for the received packet. */
2767 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2769 /* No interface? Something fishy... */
2771 /* Odd. A few possible reasons here:
2773 * - We deleted an interface but there are still a few packets
2774 * queued up from it.
2776 * - Someone externally added an interface (e.g. with "ovs-dpctl
2777 * add-if") that we don't know about.
2779 * - Packet arrived on the local port but the local port is not
2780 * one of our bridge ports.
2782 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2784 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2785 "interface %"PRIu16, br->name, flow->in_port);
2791 *in_portp = in_port = in_iface->port;
2792 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2797 /* Drop frames for reserved multicast addresses. */
2798 if (eth_addr_is_reserved(flow->dl_dst)) {
2802 /* Drop frames on ports reserved for mirroring. */
2803 if (in_port->is_mirror_output_port) {
2805 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2806 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2807 "%s, which is reserved exclusively for mirroring",
2808 br->name, in_port->name);
2813 /* When using LACP, do not accept packets from disabled interfaces. */
2814 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2818 /* Packets received on non-LACP bonds need special attention to avoid
2820 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2821 struct mac_entry *mac;
2823 if (eth_addr_is_multicast(flow->dl_dst)) {
2824 *tags |= port_get_active_iface_tag(in_port);
2825 if (in_port->active_iface != in_iface) {
2826 /* Drop all multicast packets on inactive slaves. */
2831 /* Drop all packets for which we have learned a different input
2832 * port, because we probably sent the packet on one slave and got
2833 * it back on the other. Gratuitous ARP packets are an exception
2834 * to this rule: the host has moved to another switch. The exception
2835 * to the exception is if we locked the learning table to avoid
2836 * reflections on bond slaves. If this is the case, just drop the
2838 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2839 if (mac && mac->port.p != in_port &&
2840 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2848 /* If the composed actions may be applied to any packet in the given 'flow',
2849 * returns true. Otherwise, the actions should only be applied to 'packet', or
2850 * not at all, if 'packet' was NULL. */
2852 process_flow(struct bridge *br, const struct flow *flow,
2853 const struct ofpbuf *packet, struct ofpbuf *actions,
2854 tag_type *tags, uint16_t *nf_output_iface)
2856 struct port *in_port;
2857 struct port *out_port;
2858 struct mac_entry *mac;
2861 /* Check whether we should drop packets in this flow. */
2862 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2867 /* Learn source MAC (but don't try to learn from revalidation). */
2869 update_learning_table(br, flow, vlan, in_port);
2872 /* Determine output port. */
2873 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2875 out_port = mac->port.p;
2876 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2877 /* If we are revalidating but don't have a learning entry then
2878 * eject the flow. Installing a flow that floods packets opens
2879 * up a window of time where we could learn from a packet reflected
2880 * on a bond and blackhole packets before the learning table is
2881 * updated to reflect the correct port. */
2884 out_port = FLOOD_PORT;
2887 /* Don't send packets out their input ports. */
2888 if (in_port == out_port) {
2894 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2902 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2903 struct ofpbuf *actions, tag_type *tags,
2904 uint16_t *nf_output_iface, void *br_)
2906 struct bridge *br = br_;
2908 COVERAGE_INC(bridge_process_flow);
2909 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2913 bridge_special_ofhook_cb(const struct flow *flow,
2914 const struct ofpbuf *packet, void *br_)
2916 struct iface *iface;
2917 struct bridge *br = br_;
2919 iface = iface_from_dp_ifidx(br, flow->in_port);
2921 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2923 if (iface && iface->port->lacp && packet) {
2924 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2927 COVERAGE_INC(bridge_process_lacp);
2928 lacp_process_pdu(iface->port->lacp, iface, pdu);
2938 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2939 const struct nlattr *actions,
2941 uint64_t n_bytes, void *br_)
2943 struct bridge *br = br_;
2944 const struct nlattr *a;
2945 struct port *in_port;
2950 /* Feed information from the active flows back into the learning table to
2951 * ensure that table is always in sync with what is actually flowing
2952 * through the datapath.
2954 * We test that 'tags' is nonzero to ensure that only flows that include an
2955 * OFPP_NORMAL action are used for learning. This works because
2956 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2957 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2958 update_learning_table(br, flow, vlan, in_port);
2961 /* Account for bond slave utilization. */
2962 if (!br->has_bonded_ports) {
2965 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2966 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
2967 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2968 if (out_port && out_port->n_ifaces >= 2 &&
2969 out_port->bond_mode != BM_AB) {
2970 uint16_t vlan = (flow->vlan_tci
2971 ? vlan_tci_to_vid(flow->vlan_tci)
2973 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
2974 e->tx_bytes += n_bytes;
2981 bridge_account_checkpoint_ofhook_cb(void *br_)
2983 struct bridge *br = br_;
2987 if (!br->has_bonded_ports) {
2992 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2993 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
2994 && now >= port->bond_next_rebalance) {
2995 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2996 bond_rebalance_port(port);
3001 static struct ofhooks bridge_ofhooks = {
3002 bridge_normal_ofhook_cb,
3003 bridge_special_ofhook_cb,
3004 bridge_account_flow_ofhook_cb,
3005 bridge_account_checkpoint_ofhook_cb,
3008 /* Bonding functions. */
3010 /* Statistics for a single interface on a bonded port, used for load-based
3011 * bond rebalancing. */
3012 struct slave_balance {
3013 struct iface *iface; /* The interface. */
3014 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3016 /* All the "bond_entry"s that are assigned to this interface, in order of
3017 * increasing tx_bytes. */
3018 struct bond_entry **hashes;
3023 bond_mode_to_string(enum bond_mode bm) {
3024 static char *bm_slb = "balance-slb";
3025 static char *bm_ab = "active-backup";
3026 static char *bm_tcp = "balance-tcp";
3029 case BM_SLB: return bm_slb;
3030 case BM_AB: return bm_ab;
3031 case BM_TCP: return bm_tcp;
3038 /* Sorts pointers to pointers to bond_entries in ascending order by the
3039 * interface to which they are assigned, and within a single interface in
3040 * ascending order of bytes transmitted. */
3042 compare_bond_entries(const void *a_, const void *b_)
3044 const struct bond_entry *const *ap = a_;
3045 const struct bond_entry *const *bp = b_;
3046 const struct bond_entry *a = *ap;
3047 const struct bond_entry *b = *bp;
3048 if (a->iface != b->iface) {
3049 return a->iface > b->iface ? 1 : -1;
3050 } else if (a->tx_bytes != b->tx_bytes) {
3051 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3057 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3058 * *descending* order by number of bytes transmitted. */
3060 compare_slave_balance(const void *a_, const void *b_)
3062 const struct slave_balance *a = a_;
3063 const struct slave_balance *b = b_;
3064 if (a->iface->enabled != b->iface->enabled) {
3065 return a->iface->enabled ? -1 : 1;
3066 } else if (a->tx_bytes != b->tx_bytes) {
3067 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3074 swap_bals(struct slave_balance *a, struct slave_balance *b)
3076 struct slave_balance tmp = *a;
3081 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3082 * given that 'p' (and only 'p') might be in the wrong location.
3084 * This function invalidates 'p', since it might now be in a different memory
3087 resort_bals(struct slave_balance *p,
3088 struct slave_balance bals[], size_t n_bals)
3091 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3092 swap_bals(p, p - 1);
3094 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3095 swap_bals(p, p + 1);
3101 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3103 if (VLOG_IS_DBG_ENABLED()) {
3104 struct ds ds = DS_EMPTY_INITIALIZER;
3105 const struct slave_balance *b;
3107 for (b = bals; b < bals + n_bals; b++) {
3111 ds_put_char(&ds, ',');
3113 ds_put_format(&ds, " %s %"PRIu64"kB",
3114 b->iface->name, b->tx_bytes / 1024);
3116 if (!b->iface->enabled) {
3117 ds_put_cstr(&ds, " (disabled)");
3119 if (b->n_hashes > 0) {
3120 ds_put_cstr(&ds, " (");
3121 for (i = 0; i < b->n_hashes; i++) {
3122 const struct bond_entry *e = b->hashes[i];
3124 ds_put_cstr(&ds, " + ");
3126 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3127 e - port->bond_hash, e->tx_bytes / 1024);
3129 ds_put_cstr(&ds, ")");
3132 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3137 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3139 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3142 struct bond_entry *hash = from->hashes[hash_idx];
3143 struct port *port = from->iface->port;
3144 uint64_t delta = hash->tx_bytes;
3146 assert(port->bond_mode != BM_AB);
3148 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3149 "from %s to %s (now carrying %"PRIu64"kB and "
3150 "%"PRIu64"kB load, respectively)",
3151 port->name, delta / 1024, hash - port->bond_hash,
3152 from->iface->name, to->iface->name,
3153 (from->tx_bytes - delta) / 1024,
3154 (to->tx_bytes + delta) / 1024);
3156 /* Delete element from from->hashes.
3158 * We don't bother to add the element to to->hashes because not only would
3159 * it require more work, the only purpose it would be to allow that hash to
3160 * be migrated to another slave in this rebalancing run, and there is no
3161 * point in doing that. */
3162 if (hash_idx == 0) {
3165 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3166 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3170 /* Shift load away from 'from' to 'to'. */
3171 from->tx_bytes -= delta;
3172 to->tx_bytes += delta;
3174 /* Arrange for flows to be revalidated. */
3175 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3176 hash->iface = to->iface;
3177 hash->tag = tag_create_random();
3181 bond_rebalance_port(struct port *port)
3183 struct slave_balance *bals;
3185 struct bond_entry *hashes[BOND_MASK + 1];
3186 struct slave_balance *b, *from, *to;
3187 struct bond_entry *e;
3188 struct iface *iface;
3191 assert(port->bond_mode != BM_AB);
3193 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3194 * descending order of tx_bytes, so that bals[0] represents the most
3195 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3198 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3199 * array for each slave_balance structure, we sort our local array of
3200 * hashes in order by slave, so that all of the hashes for a given slave
3201 * become contiguous in memory, and then we point each 'hashes' members of
3202 * a slave_balance structure to the start of a contiguous group. */
3203 n_bals = port->n_ifaces;
3204 b = bals = xmalloc(n_bals * sizeof *bals);
3205 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3212 assert(b == &bals[n_bals]);
3213 for (i = 0; i <= BOND_MASK; i++) {
3214 hashes[i] = &port->bond_hash[i];
3216 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3217 for (i = 0; i <= BOND_MASK; i++) {
3223 for (b = bals; b < &bals[n_bals]; b++) {
3224 if (b->iface == e->iface) {
3225 b->tx_bytes += e->tx_bytes;
3227 b->hashes = &hashes[i];
3234 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3235 log_bals(bals, n_bals, port);
3237 /* Discard slaves that aren't enabled (which were sorted to the back of the
3238 * array earlier). */
3239 while (!bals[n_bals - 1].iface->enabled) {
3246 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3247 to = &bals[n_bals - 1];
3248 for (from = bals; from < to; ) {
3249 uint64_t overload = from->tx_bytes - to->tx_bytes;
3250 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3251 /* The extra load on 'from' (and all less-loaded slaves), compared
3252 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3253 * it is less than ~1Mbps. No point in rebalancing. */
3255 } else if (from->n_hashes == 1) {
3256 /* 'from' only carries a single MAC hash, so we can't shift any
3257 * load away from it, even though we want to. */
3260 /* 'from' is carrying significantly more load than 'to', and that
3261 * load is split across at least two different hashes. Pick a hash
3262 * to migrate to 'to' (the least-loaded slave), given that doing so
3263 * must decrease the ratio of the load on the two slaves by at
3266 * The sort order we use means that we prefer to shift away the
3267 * smallest hashes instead of the biggest ones. There is little
3268 * reason behind this decision; we could use the opposite sort
3269 * order to shift away big hashes ahead of small ones. */
3272 for (i = 0; i < from->n_hashes; i++) {
3273 double old_ratio, new_ratio;
3274 uint64_t delta = from->hashes[i]->tx_bytes;
3276 if (delta == 0 || from->tx_bytes - delta == 0) {
3277 /* Pointless move. */
3281 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3283 if (to->tx_bytes == 0) {
3284 /* Nothing on the new slave, move it. */
3288 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3289 new_ratio = (double)(from->tx_bytes - delta) /
3290 (to->tx_bytes + delta);
3292 if (new_ratio == 0) {
3293 /* Should already be covered but check to prevent division
3298 if (new_ratio < 1) {
3299 new_ratio = 1 / new_ratio;
3302 if (old_ratio - new_ratio > 0.1) {
3303 /* Would decrease the ratio, move it. */
3307 if (i < from->n_hashes) {
3308 bond_shift_load(from, to, i);
3310 /* If the result of the migration changed the relative order of
3311 * 'from' and 'to' swap them back to maintain invariants. */
3312 if (order_swapped) {
3313 swap_bals(from, to);
3316 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3317 * point to different slave_balance structures. It is only
3318 * valid to do these two operations in a row at all because we
3319 * know that 'from' will not move past 'to' and vice versa. */
3320 resort_bals(from, bals, n_bals);
3321 resort_bals(to, bals, n_bals);
3328 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3329 * historical data to decay to <1% in 7 rebalancing runs. */
3330 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3342 bond_send_learning_packets(struct port *port)
3344 struct bridge *br = port->bridge;
3345 struct mac_entry *e;
3346 struct ofpbuf packet;
3347 int error, n_packets, n_errors;
3349 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3353 ofpbuf_init(&packet, 128);
3354 error = n_packets = n_errors = 0;
3355 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3361 if (e->port.p == port) {
3365 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3367 flow_extract(&packet, 0, ODPP_NONE, &flow);
3369 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3375 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3381 ofpbuf_uninit(&packet);
3384 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3385 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3386 "packets, last error was: %s",
3387 port->name, n_errors, n_packets, strerror(error));
3389 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3390 port->name, n_packets);
3394 /* Bonding unixctl user interface functions. */
3397 bond_unixctl_list(struct unixctl_conn *conn,
3398 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3400 struct ds ds = DS_EMPTY_INITIALIZER;
3401 const struct bridge *br;
3403 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3405 LIST_FOR_EACH (br, node, &all_bridges) {
3408 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3409 if (port->n_ifaces > 1) {
3410 struct iface *iface;
3412 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3413 bond_mode_to_string(port->bond_mode));
3414 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3415 if (&iface->port_elem != list_front(&port->ifaces)) {
3416 ds_put_cstr(&ds, ", ");
3418 ds_put_cstr(&ds, iface->name);
3420 ds_put_char(&ds, '\n');
3424 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3428 static struct port *
3429 bond_find(const char *name)
3431 const struct bridge *br;
3433 LIST_FOR_EACH (br, node, &all_bridges) {
3436 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3437 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3446 bond_unixctl_show(struct unixctl_conn *conn,
3447 const char *args, void *aux OVS_UNUSED)
3449 struct ds ds = DS_EMPTY_INITIALIZER;
3450 const struct port *port;
3451 struct iface *iface;
3453 port = bond_find(args);
3455 unixctl_command_reply(conn, 501, "no such bond");
3459 ds_put_format(&ds, "bond_mode: %s\n",
3460 bond_mode_to_string(port->bond_mode));
3463 ds_put_format(&ds, "lacp: %s\n",
3464 port->lacp_active ? "active" : "passive");
3466 ds_put_cstr(&ds, "lacp: off\n");
3469 if (port->bond_mode != BM_AB) {
3470 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3471 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3475 ds_put_format(&ds, "bond-detect-mode: %s\n",
3476 port->monitor ? "carrier" : "miimon");
3478 if (!port->monitor) {
3479 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3480 port->miimon_interval);
3483 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3484 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3486 if (port->bond_mode != BM_AB) {
3487 ds_put_format(&ds, "next rebalance: %lld ms\n",
3488 port->bond_next_rebalance - time_msec());
3491 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3492 struct bond_entry *be;
3496 ds_put_format(&ds, "\nslave %s: %s\n",
3497 iface->name, iface->enabled ? "enabled" : "disabled");
3498 if (iface == port->active_iface) {
3499 ds_put_cstr(&ds, "\tactive slave\n");
3501 if (iface->delay_expires != LLONG_MAX) {
3502 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3503 iface->enabled ? "downdelay" : "updelay",
3504 iface->delay_expires - time_msec());
3507 if (port->bond_mode == BM_AB) {
3512 memset(&flow, 0, sizeof flow);
3513 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3514 int hash = be - port->bond_hash;
3515 struct mac_entry *me;
3517 if (be->iface != iface) {
3521 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3522 hash, be->tx_bytes / 1024);
3524 if (port->bond_mode != BM_SLB) {
3529 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3533 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3534 if (bond_hash_src(me->mac, me->vlan) == hash
3535 && me->port.p != port
3536 && choose_output_iface(port, &flow, me->vlan,
3538 && dp_ifidx == iface->dp_ifidx)
3540 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3541 ETH_ADDR_ARGS(me->mac));
3546 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3551 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3552 void *aux OVS_UNUSED)
3554 char *args = (char *) args_;
3555 char *save_ptr = NULL;
3556 char *bond_s, *hash_s, *slave_s;
3558 struct iface *iface;
3559 struct bond_entry *entry;
3562 bond_s = strtok_r(args, " ", &save_ptr);
3563 hash_s = strtok_r(NULL, " ", &save_ptr);
3564 slave_s = strtok_r(NULL, " ", &save_ptr);
3566 unixctl_command_reply(conn, 501,
3567 "usage: bond/migrate BOND HASH SLAVE");
3571 port = bond_find(bond_s);
3573 unixctl_command_reply(conn, 501, "no such bond");
3577 if (port->bond_mode != BM_SLB) {
3578 unixctl_command_reply(conn, 501, "not an SLB bond");
3582 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3583 hash = atoi(hash_s) & BOND_MASK;
3585 unixctl_command_reply(conn, 501, "bad hash");
3589 iface = port_lookup_iface(port, slave_s);
3591 unixctl_command_reply(conn, 501, "no such slave");
3595 if (!iface->enabled) {
3596 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3600 entry = &port->bond_hash[hash];
3601 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3602 entry->iface = iface;
3603 entry->tag = tag_create_random();
3604 unixctl_command_reply(conn, 200, "migrated");
3608 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3609 void *aux OVS_UNUSED)
3611 char *args = (char *) args_;
3612 char *save_ptr = NULL;
3613 char *bond_s, *slave_s;
3615 struct iface *iface;
3617 bond_s = strtok_r(args, " ", &save_ptr);
3618 slave_s = strtok_r(NULL, " ", &save_ptr);
3620 unixctl_command_reply(conn, 501,
3621 "usage: bond/set-active-slave BOND SLAVE");
3625 port = bond_find(bond_s);
3627 unixctl_command_reply(conn, 501, "no such bond");
3631 iface = port_lookup_iface(port, slave_s);
3633 unixctl_command_reply(conn, 501, "no such slave");
3637 if (!iface->enabled) {
3638 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3642 if (port->active_iface != iface) {
3643 ofproto_revalidate(port->bridge->ofproto,
3644 port_get_active_iface_tag(port));
3645 port->active_iface = iface;
3646 VLOG_INFO("port %s: active interface is now %s",
3647 port->name, iface->name);
3648 bond_send_learning_packets(port);
3649 unixctl_command_reply(conn, 200, "done");
3651 unixctl_command_reply(conn, 200, "no change");
3656 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3658 char *args = (char *) args_;
3659 char *save_ptr = NULL;
3660 char *bond_s, *slave_s;
3662 struct iface *iface;
3664 bond_s = strtok_r(args, " ", &save_ptr);
3665 slave_s = strtok_r(NULL, " ", &save_ptr);
3667 unixctl_command_reply(conn, 501,
3668 "usage: bond/enable/disable-slave BOND SLAVE");
3672 port = bond_find(bond_s);
3674 unixctl_command_reply(conn, 501, "no such bond");
3678 iface = port_lookup_iface(port, slave_s);
3680 unixctl_command_reply(conn, 501, "no such slave");
3684 bond_enable_slave(iface, enable);
3685 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3689 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3690 void *aux OVS_UNUSED)
3692 enable_slave(conn, args, true);
3696 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3697 void *aux OVS_UNUSED)
3699 enable_slave(conn, args, false);
3703 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3704 void *aux OVS_UNUSED)
3706 char *args = (char *) args_;
3707 uint8_t mac[ETH_ADDR_LEN];
3711 char *mac_s, *vlan_s;
3712 char *save_ptr = NULL;
3714 mac_s = strtok_r(args, " ", &save_ptr);
3715 vlan_s = strtok_r(NULL, " ", &save_ptr);
3718 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3719 unixctl_command_reply(conn, 501, "invalid vlan");
3723 vlan = OFP_VLAN_NONE;
3726 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3727 == ETH_ADDR_SCAN_COUNT) {
3728 hash = bond_hash_src(mac, vlan);
3730 hash_cstr = xasprintf("%u", hash);
3731 unixctl_command_reply(conn, 200, hash_cstr);
3734 unixctl_command_reply(conn, 501, "invalid mac");
3741 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3742 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3743 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3744 unixctl_command_register("bond/set-active-slave",
3745 bond_unixctl_set_active_slave, NULL);
3746 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3748 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3750 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3753 /* Port functions. */
3756 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3758 struct iface *iface = aux;
3759 uint8_t ea[ETH_ADDR_LEN];
3762 error = netdev_get_etheraddr(iface->netdev, ea);
3764 struct ofpbuf packet;
3765 struct lacp_pdu *packet_pdu;
3767 ofpbuf_init(&packet, 0);
3768 packet_pdu = compose_packet(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3769 sizeof *packet_pdu);
3770 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3771 ofproto_send_packet(iface->port->bridge->ofproto,
3772 iface->dp_ifidx, 0, &packet);
3773 ofpbuf_uninit(&packet);
3775 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3776 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3777 "(%s)", iface->name, strerror(error));
3782 port_run(struct port *port)
3784 if (port->monitor) {
3787 /* Track carrier going up and down on interfaces. */
3788 while (!netdev_monitor_poll(port->monitor, &devname)) {
3789 struct iface *iface;
3791 iface = port_lookup_iface(port, devname);
3793 iface_update_carrier(iface);
3797 } else if (time_msec() >= port->miimon_next_update) {
3798 struct iface *iface;
3800 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3801 iface_update_carrier(iface);
3803 port->miimon_next_update = time_msec() + port->miimon_interval;
3807 struct iface *iface;
3809 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3810 lacp_slave_enable(port->lacp, iface, iface->enabled);
3813 lacp_run(port->lacp, lacp_send_pdu_cb);
3820 port_wait(struct port *port)
3822 if (port->monitor) {
3823 netdev_monitor_poll_wait(port->monitor);
3825 poll_timer_wait_until(port->miimon_next_update);
3829 lacp_wait(port->lacp);
3835 static struct port *
3836 port_create(struct bridge *br, const char *name)
3840 port = xzalloc(sizeof *port);
3843 port->trunks = NULL;
3844 port->name = xstrdup(name);
3845 port->active_iface = NULL;
3846 list_init(&port->ifaces);
3848 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3850 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3857 get_port_other_config(const struct ovsrec_port *port, const char *key,
3858 const char *default_value)
3862 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3864 return value ? value : default_value;
3868 get_interface_other_config(const struct ovsrec_interface *iface,
3869 const char *key, const char *default_value)
3873 value = get_ovsrec_key_value(&iface->header_,
3874 &ovsrec_interface_col_other_config, key);
3875 return value ? value : default_value;
3879 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3881 struct iface *iface, *next;
3882 struct shash new_ifaces;
3885 /* Collect list of new interfaces. */
3886 shash_init(&new_ifaces);
3887 for (i = 0; i < cfg->n_interfaces; i++) {
3888 const char *name = cfg->interfaces[i]->name;
3889 shash_add_once(&new_ifaces, name, NULL);
3892 /* Get rid of deleted interfaces. */
3893 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
3894 if (!shash_find(&new_ifaces, iface->name)) {
3895 iface_destroy(iface);
3899 shash_destroy(&new_ifaces);
3902 /* Expires all MAC learning entries associated with 'port' and forces ofproto
3903 * to revalidate every flow. */
3905 port_flush_macs(struct port *port)
3907 struct bridge *br = port->bridge;
3908 struct mac_learning *ml = br->ml;
3909 struct mac_entry *mac, *next_mac;
3912 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
3913 if (mac->port.p == port) {
3914 mac_learning_expire(ml, mac);
3920 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3922 const char *detect_mode;
3923 struct shash new_ifaces;
3924 long long int next_rebalance, miimon_next_update, lacp_priority;
3925 bool need_flush = false;
3926 unsigned long *trunks;
3932 /* Update settings. */
3933 port->updelay = cfg->bond_updelay;
3934 if (port->updelay < 0) {
3937 port->downdelay = cfg->bond_downdelay;
3938 if (port->downdelay < 0) {
3939 port->downdelay = 0;
3941 port->bond_rebalance_interval = atoi(
3942 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3943 if (port->bond_rebalance_interval < 1000) {
3944 port->bond_rebalance_interval = 1000;
3946 next_rebalance = time_msec() + port->bond_rebalance_interval;
3947 if (port->bond_next_rebalance > next_rebalance) {
3948 port->bond_next_rebalance = next_rebalance;
3951 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3954 netdev_monitor_destroy(port->monitor);
3955 port->monitor = NULL;
3957 if (strcmp(detect_mode, "miimon")) {
3958 port->monitor = netdev_monitor_create();
3960 if (strcmp(detect_mode, "carrier")) {
3961 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
3962 "defaulting to carrier", port->name, detect_mode);
3966 port->miimon_interval = atoi(
3967 get_port_other_config(cfg, "bond-miimon-interval", "200"));
3968 if (port->miimon_interval < 100) {
3969 port->miimon_interval = 100;
3971 miimon_next_update = time_msec() + port->miimon_interval;
3972 if (port->miimon_next_update > miimon_next_update) {
3973 port->miimon_next_update = miimon_next_update;
3976 if (!port->cfg->bond_mode ||
3977 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
3978 port->bond_mode = BM_SLB;
3979 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
3980 port->bond_mode = BM_AB;
3981 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
3982 port->bond_mode = BM_TCP;
3984 port->bond_mode = BM_SLB;
3985 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
3986 port->name, port->cfg->bond_mode,
3987 bond_mode_to_string(port->bond_mode));
3990 /* Add new interfaces and update 'cfg' member of existing ones. */
3991 shash_init(&new_ifaces);
3992 for (i = 0; i < cfg->n_interfaces; i++) {
3993 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3994 struct iface *iface;
3996 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3997 VLOG_WARN("port %s: %s specified twice as port interface",
3998 port->name, if_cfg->name);
3999 iface_set_ofport(if_cfg, -1);
4003 iface = iface_lookup(port->bridge, if_cfg->name);
4005 if (iface->port != port) {
4006 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4008 port->bridge->name, if_cfg->name, iface->port->name);
4011 iface->cfg = if_cfg;
4013 iface = iface_create(port, if_cfg);
4016 /* Determine interface type. The local port always has type
4017 * "internal". Other ports take their type from the database and
4018 * default to "system" if none is specified. */
4019 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4020 : if_cfg->type[0] ? if_cfg->type
4024 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4027 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4028 iface->lacp_priority = UINT16_MAX;
4030 iface->lacp_priority = lacp_priority;
4033 shash_destroy(&new_ifaces);
4035 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4039 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4041 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4042 /* Prefer bondable links if unspecified. */
4043 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4045 port->lacp_priority = lacp_priority;
4048 if (!port->cfg->lacp) {
4049 /* XXX when LACP implementation has been sufficiently tested, enable by
4050 * default and make active on bonded ports. */
4051 lacp_destroy(port->lacp);
4053 } else if (!strcmp(port->cfg->lacp, "off")) {
4054 lacp_destroy(port->lacp);
4056 } else if (!strcmp(port->cfg->lacp, "active")) {
4058 port->lacp = lacp_create();
4060 port->lacp_active = true;
4061 } else if (!strcmp(port->cfg->lacp, "passive")) {
4063 port->lacp = lacp_create();
4065 port->lacp_active = false;
4067 VLOG_WARN("port %s: unknown LACP mode %s",
4068 port->name, port->cfg->lacp);
4069 lacp_destroy(port->lacp);
4076 if (port->n_ifaces < 2) {
4078 if (vlan >= 0 && vlan <= 4095) {
4079 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4084 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4085 * they even work as-is. But they have not been tested. */
4086 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4090 if (port->vlan != vlan) {
4095 /* Get trunked VLANs. */
4097 if (vlan < 0 && cfg->n_trunks) {
4100 trunks = bitmap_allocate(4096);
4102 for (i = 0; i < cfg->n_trunks; i++) {
4103 int trunk = cfg->trunks[i];
4105 bitmap_set1(trunks, trunk);
4111 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4112 port->name, cfg->n_trunks);
4114 if (n_errors == cfg->n_trunks) {
4115 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4117 bitmap_free(trunks);
4120 } else if (vlan >= 0 && cfg->n_trunks) {
4121 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4125 ? port->trunks != NULL
4126 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4129 bitmap_free(port->trunks);
4130 port->trunks = trunks;
4133 port_flush_macs(port);
4138 port_destroy(struct port *port)
4141 struct bridge *br = port->bridge;
4142 struct iface *iface, *next;
4145 for (i = 0; i < MAX_MIRRORS; i++) {
4146 struct mirror *m = br->mirrors[i];
4147 if (m && m->out_port == port) {
4152 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4153 iface_destroy(iface);
4156 hmap_remove(&br->ports, &port->hmap_node);
4158 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4160 port_flush_macs(port);
4162 lacp_destroy(port->lacp);
4163 netdev_monitor_destroy(port->monitor);
4164 bitmap_free(port->trunks);
4165 free(port->bond_hash);
4171 static struct port *
4172 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4174 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4175 return iface ? iface->port : NULL;
4178 static struct port *
4179 port_lookup(const struct bridge *br, const char *name)
4183 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4185 if (!strcmp(port->name, name)) {
4192 static struct iface *
4193 port_lookup_iface(const struct port *port, const char *name)
4195 struct iface *iface = iface_lookup(port->bridge, name);
4196 return iface && iface->port == port ? iface : NULL;
4200 port_update_lacp(struct port *port)
4203 struct iface *iface;
4205 lacp_configure(port->lacp, port->name,
4206 port->bridge->ea, port->lacp_priority,
4207 port->lacp_active, port->lacp_fast);
4209 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4210 lacp_slave_register(port->lacp, iface, iface->name,
4211 iface->dp_ifidx, iface->lacp_priority);
4217 port_update_bonding(struct port *port)
4219 if (port->n_ifaces < 2) {
4220 /* Not a bonded port. */
4221 free(port->bond_hash);
4222 port->bond_hash = NULL;
4223 port->bond_fake_iface = false;
4224 port->active_iface = NULL;
4225 port->no_ifaces_tag = 0;
4229 if (port->bond_mode != BM_AB && !port->bond_hash) {
4230 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4231 for (i = 0; i <= BOND_MASK; i++) {
4232 struct bond_entry *e = &port->bond_hash[i];
4236 port->bond_next_rebalance
4237 = time_msec() + port->bond_rebalance_interval;
4238 } else if (port->bond_mode == BM_AB) {
4239 free(port->bond_hash);
4240 port->bond_hash = NULL;
4243 if (!port->no_ifaces_tag) {
4244 port->no_ifaces_tag = tag_create_random();
4247 if (!port->active_iface) {
4248 bond_choose_active_iface(port);
4251 port->bond_fake_iface = port->cfg->bond_fake_iface;
4252 if (port->bond_fake_iface) {
4253 port->bond_next_fake_iface_update = time_msec();
4259 /* Interface functions. */
4261 static struct iface *
4262 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4264 struct bridge *br = port->bridge;
4265 struct iface *iface;
4266 char *name = if_cfg->name;
4268 iface = xzalloc(sizeof *iface);
4270 iface->name = xstrdup(name);
4271 iface->dp_ifidx = -1;
4272 iface->tag = tag_create_random();
4273 iface->delay_expires = LLONG_MAX;
4274 iface->netdev = NULL;
4275 iface->cfg = if_cfg;
4277 shash_add_assert(&br->iface_by_name, iface->name, iface);
4279 list_push_back(&port->ifaces, &iface->port_elem);
4282 if (port->n_ifaces > 1) {
4283 br->has_bonded_ports = true;
4286 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4294 iface_destroy(struct iface *iface)
4297 struct port *port = iface->port;
4298 struct bridge *br = port->bridge;
4299 bool del_active = port->active_iface == iface;
4301 if (port->bond_hash) {
4302 struct bond_entry *e;
4303 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4304 if (e->iface == iface) {
4310 if (iface->port->lacp) {
4311 lacp_slave_unregister(iface->port->lacp, iface);
4314 if (port->monitor && iface->netdev) {
4315 netdev_monitor_remove(port->monitor, iface->netdev);
4318 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4320 if (iface->dp_ifidx >= 0) {
4321 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4324 list_remove(&iface->port_elem);
4327 netdev_close(iface->netdev);
4330 bond_choose_active_iface(port);
4331 bond_send_learning_packets(port);
4337 bridge_flush(port->bridge);
4341 static struct iface *
4342 iface_lookup(const struct bridge *br, const char *name)
4344 return shash_find_data(&br->iface_by_name, name);
4347 static struct iface *
4348 iface_find(const char *name)
4350 const struct bridge *br;
4352 LIST_FOR_EACH (br, node, &all_bridges) {
4353 struct iface *iface = iface_lookup(br, name);
4362 static struct iface *
4363 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4365 struct iface *iface;
4367 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4368 hash_int(dp_ifidx, 0), &br->ifaces) {
4369 if (iface->dp_ifidx == dp_ifidx) {
4376 /* Set Ethernet address of 'iface', if one is specified in the configuration
4379 iface_set_mac(struct iface *iface)
4381 uint8_t ea[ETH_ADDR_LEN];
4383 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4384 if (eth_addr_is_multicast(ea)) {
4385 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4387 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4388 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4389 iface->name, iface->name);
4391 int error = netdev_set_etheraddr(iface->netdev, ea);
4393 VLOG_ERR("interface %s: setting MAC failed (%s)",
4394 iface->name, strerror(error));
4400 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4402 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4405 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4409 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4411 * The value strings in '*shash' are taken directly from values[], not copied,
4412 * so the caller should not modify or free them. */
4414 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4415 struct shash *shash)
4420 for (i = 0; i < n; i++) {
4421 shash_add(shash, keys[i], values[i]);
4425 /* Creates 'keys' and 'values' arrays from 'shash'.
4427 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4428 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4429 * are populated with with strings taken directly from 'shash' and thus have
4430 * the same ownership of the key-value pairs in shash.
4433 shash_to_ovs_idl_map(struct shash *shash,
4434 char ***keys, char ***values, size_t *n)
4438 struct shash_node *sn;
4440 count = shash_count(shash);
4442 k = xmalloc(count * sizeof *k);
4443 v = xmalloc(count * sizeof *v);
4446 SHASH_FOR_EACH(sn, shash) {
4457 struct iface_delete_queues_cbdata {
4458 struct netdev *netdev;
4459 const struct ovsdb_datum *queues;
4463 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4465 union ovsdb_atom atom;
4467 atom.integer = target;
4468 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4472 iface_delete_queues(unsigned int queue_id,
4473 const struct shash *details OVS_UNUSED, void *cbdata_)
4475 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4477 if (!queue_ids_include(cbdata->queues, queue_id)) {
4478 netdev_delete_queue(cbdata->netdev, queue_id);
4483 iface_update_carrier(struct iface *iface)
4485 bool carrier = iface_get_carrier(iface);
4486 if (carrier == iface->up) {
4490 iface->up = carrier;
4491 if (iface->port->lacp) {
4492 lacp_slave_carrier_changed(iface->port->lacp, iface);
4497 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4499 if (!qos || qos->type[0] == '\0') {
4500 netdev_set_qos(iface->netdev, NULL, NULL);
4502 struct iface_delete_queues_cbdata cbdata;
4503 struct shash details;
4506 /* Configure top-level Qos for 'iface'. */
4507 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4508 qos->n_other_config, &details);
4509 netdev_set_qos(iface->netdev, qos->type, &details);
4510 shash_destroy(&details);
4512 /* Deconfigure queues that were deleted. */
4513 cbdata.netdev = iface->netdev;
4514 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4516 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4518 /* Configure queues for 'iface'. */
4519 for (i = 0; i < qos->n_queues; i++) {
4520 const struct ovsrec_queue *queue = qos->value_queues[i];
4521 unsigned int queue_id = qos->key_queues[i];
4523 shash_from_ovs_idl_map(queue->key_other_config,
4524 queue->value_other_config,
4525 queue->n_other_config, &details);
4526 netdev_set_queue(iface->netdev, queue_id, &details);
4527 shash_destroy(&details);
4533 iface_update_cfm(struct iface *iface)
4537 uint16_t *remote_mps;
4538 struct ovsrec_monitor *mon;
4539 uint8_t maid[CCM_MAID_LEN];
4541 mon = iface->cfg->monitor;
4544 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4548 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4549 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4553 cfm.mpid = mon->mpid;
4554 cfm.interval = mon->interval ? *mon->interval : 1000;
4556 memcpy(cfm.maid, maid, sizeof cfm.maid);
4558 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4559 for(i = 0; i < mon->n_remote_mps; i++) {
4560 remote_mps[i] = mon->remote_mps[i]->mpid;
4563 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4564 &cfm, remote_mps, mon->n_remote_mps);
4568 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4569 * how 'iface''s port is configured.
4571 * Returns true if 'iface' is up, false otherwise. */
4573 iface_get_carrier(const struct iface *iface)
4575 return (iface->port->monitor
4576 ? netdev_get_carrier(iface->netdev)
4577 : netdev_get_miimon(iface->netdev));
4580 /* Port mirroring. */
4582 static struct mirror *
4583 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4587 for (i = 0; i < MAX_MIRRORS; i++) {
4588 struct mirror *m = br->mirrors[i];
4589 if (m && uuid_equals(uuid, &m->uuid)) {
4597 mirror_reconfigure(struct bridge *br)
4599 unsigned long *rspan_vlans;
4603 /* Get rid of deleted mirrors. */
4604 for (i = 0; i < MAX_MIRRORS; i++) {
4605 struct mirror *m = br->mirrors[i];
4607 const struct ovsdb_datum *mc;
4608 union ovsdb_atom atom;
4610 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4611 atom.uuid = br->mirrors[i]->uuid;
4612 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4618 /* Add new mirrors and reconfigure existing ones. */
4619 for (i = 0; i < br->cfg->n_mirrors; i++) {
4620 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4621 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4623 mirror_reconfigure_one(m, cfg);
4625 mirror_create(br, cfg);
4629 /* Update port reserved status. */
4630 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4631 port->is_mirror_output_port = false;
4633 for (i = 0; i < MAX_MIRRORS; i++) {
4634 struct mirror *m = br->mirrors[i];
4635 if (m && m->out_port) {
4636 m->out_port->is_mirror_output_port = true;
4640 /* Update flooded vlans (for RSPAN). */
4642 if (br->cfg->n_flood_vlans) {
4643 rspan_vlans = bitmap_allocate(4096);
4645 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4646 int64_t vlan = br->cfg->flood_vlans[i];
4647 if (vlan >= 0 && vlan < 4096) {
4648 bitmap_set1(rspan_vlans, vlan);
4649 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4652 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4657 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4659 mac_learning_flush(br->ml);
4664 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4669 for (i = 0; ; i++) {
4670 if (i >= MAX_MIRRORS) {
4671 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4672 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4675 if (!br->mirrors[i]) {
4680 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4682 mac_learning_flush(br->ml);
4684 br->mirrors[i] = m = xzalloc(sizeof *m);
4687 m->name = xstrdup(cfg->name);
4688 shash_init(&m->src_ports);
4689 shash_init(&m->dst_ports);
4695 mirror_reconfigure_one(m, cfg);
4699 mirror_destroy(struct mirror *m)
4702 struct bridge *br = m->bridge;
4705 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4706 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4707 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4710 shash_destroy(&m->src_ports);
4711 shash_destroy(&m->dst_ports);
4714 m->bridge->mirrors[m->idx] = NULL;
4719 mac_learning_flush(br->ml);
4724 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4725 struct shash *names)
4729 for (i = 0; i < n_ports; i++) {
4730 const char *name = ports[i]->name;
4731 if (port_lookup(m->bridge, name)) {
4732 shash_add_once(names, name, NULL);
4734 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4735 "port %s", m->bridge->name, m->name, name);
4741 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4747 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4749 for (i = 0; i < cfg->n_select_vlan; i++) {
4750 int64_t vlan = cfg->select_vlan[i];
4751 if (vlan < 0 || vlan > 4095) {
4752 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4753 m->bridge->name, m->name, vlan);
4755 (*vlans)[n_vlans++] = vlan;
4762 vlan_is_mirrored(const struct mirror *m, int vlan)
4766 for (i = 0; i < m->n_vlans; i++) {
4767 if (m->vlans[i] == vlan) {
4775 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4779 for (i = 0; i < m->n_vlans; i++) {
4780 if (port_trunks_vlan(p, m->vlans[i])) {
4788 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4790 struct shash src_ports, dst_ports;
4791 mirror_mask_t mirror_bit;
4792 struct port *out_port;
4799 if (strcmp(cfg->name, m->name)) {
4801 m->name = xstrdup(cfg->name);
4804 /* Get output port. */
4805 if (cfg->output_port) {
4806 out_port = port_lookup(m->bridge, cfg->output_port->name);
4808 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4809 m->bridge->name, m->name);
4815 if (cfg->output_vlan) {
4816 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4817 "output vlan; ignoring output vlan",
4818 m->bridge->name, m->name);
4820 } else if (cfg->output_vlan) {
4822 out_vlan = *cfg->output_vlan;
4824 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4825 m->bridge->name, m->name);
4830 shash_init(&src_ports);
4831 shash_init(&dst_ports);
4832 if (cfg->select_all) {
4833 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4834 shash_add_once(&src_ports, port->name, NULL);
4835 shash_add_once(&dst_ports, port->name, NULL);
4840 /* Get ports, and drop duplicates and ports that don't exist. */
4841 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4843 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4846 /* Get all the vlans, and drop duplicate and invalid vlans. */
4847 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4850 /* Update mirror data. */
4851 if (!shash_equal_keys(&m->src_ports, &src_ports)
4852 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4853 || m->n_vlans != n_vlans
4854 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4855 || m->out_port != out_port
4856 || m->out_vlan != out_vlan) {
4857 bridge_flush(m->bridge);
4858 mac_learning_flush(m->bridge->ml);
4860 shash_swap(&m->src_ports, &src_ports);
4861 shash_swap(&m->dst_ports, &dst_ports);
4864 m->n_vlans = n_vlans;
4865 m->out_port = out_port;
4866 m->out_vlan = out_vlan;
4869 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4870 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4871 if (shash_find(&m->src_ports, port->name)
4874 ? port_trunks_any_mirrored_vlan(m, port)
4875 : vlan_is_mirrored(m, port->vlan)))) {
4876 port->src_mirrors |= mirror_bit;
4878 port->src_mirrors &= ~mirror_bit;
4881 if (shash_find(&m->dst_ports, port->name)) {
4882 port->dst_mirrors |= mirror_bit;
4884 port->dst_mirrors &= ~mirror_bit;
4889 shash_destroy(&src_ports);
4890 shash_destroy(&dst_ports);